diff --git a/Extras/BulletMultiThreaded/PlatformDefinitions.h b/Extras/BulletMultiThreaded/PlatformDefinitions.h
new file mode 100644
index 000000000..81566c567
--- /dev/null
+++ b/Extras/BulletMultiThreaded/PlatformDefinitions.h
@@ -0,0 +1,41 @@
+#ifndef TYPE_DEFINITIONS_H
+#define TYPE_DEFINITIONS_H
+
+#include <LinearMath/btScalar.h>
+
+///This file provides some platform/compiler checks for common definitions
+
+#ifdef WIN32
+		#if defined(__MINGW32__) || defined(__CYGWIN__) || (defined (_MSC_VER) && _MSC_VER < 1300)
+		#else
+		#endif //__MINGW32__
+
+		typedef unsigned char     uint8_t;
+		typedef unsigned long int uint64_t;
+		typedef unsigned int      uint32_t;
+		typedef unsigned short    uint16_t;
+
+		#include <malloc.h>
+		#define memalign(alignment, size) malloc(size);
+			
+#include <string.h> //memcpy
+
+		#define USE_WIN32_THREADING 1
+
+#else
+#if defined	(__CELLOS_LV2__)
+				///Playstation 3 Cell SDK
+#else
+	//non-windows systems
+
+	#define USE_PTHREADS 1
+
+#endif	//__CELLOS_LV2__
+
+	#include <stdint.h>
+	#include <stdlib.h>
+
+#endif
+
+
+#endif //TYPE_DEFINITIONS_H
\ No newline at end of file
diff --git a/Extras/BulletMultiThreaded/SpuCollisionTaskProcess.cpp b/Extras/BulletMultiThreaded/SpuCollisionTaskProcess.cpp
new file mode 100644
index 000000000..adf348b37
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuCollisionTaskProcess.cpp
@@ -0,0 +1,309 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+//#define __CELLOS_LV2__ 1
+
+//#define DEBUG_SPU_TASK_SCHEDULING 1
+
+#include "SpuLibspe2Support.h"
+#include "Win32ThreadSupport.h"
+
+//#include "SPUAssert.h"
+#include <string.h>
+
+//class OptimizedBvhNode;
+
+#include "SpuCollisionTaskProcess.h"
+#include "BulletCollision/CollisionDispatch/btCollisionObject.h"
+#include "BulletCollision/CollisionShapes/btCollisionShape.h"
+#include "BulletCollision/CollisionShapes/btConvexShape.h"
+#include "SpuLibspe2Support.h"
+#include "SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h" // for definitions processCollisionTask and createCollisionLocalStoreMemory
+
+
+#include <stdio.h>
+
+
+//#ifdef __CELLOS_LV2__???
+#ifdef USE_WIN32_THREADING
+Win32ThreadSupport gMidphaseSPU(Win32ThreadSupport::Win32ThreadConstructionInfo("collision",
+								processCollisionTask,
+								createCollisionLocalStoreMemory,
+								MIDPHASE_NUM_WORKUNIT_TASKS));
+#elif defined(USE_LIBSPE2)
+SpuLibspe2Support gMidphaseSPU(SPU_ELF_COLLISION_DETECTION,MIDPHASE_NUM_WORKUNIT_TASKS);
+#endif
+
+#ifdef __CELLOS_LV2__
+
+
+
+#endif
+
+
+SpuGatherAndProcessPairsTaskDesc g_spuGatherTaskDesc[MIDPHASE_NUM_WORKUNIT_TASKS];
+
+
+
+
+SpuCollisionTaskProcess::SpuCollisionTaskProcess()
+{
+	m_workUnitTaskBuffers = (unsigned char *)0;
+
+
+	for (int i = 0; i < MIDPHASE_NUM_WORKUNIT_TASKS; i++)
+	{
+		m_taskBusy[i] = false;
+	}
+	m_numBusyTasks = 0;
+	m_currentTask = 0;
+	m_currentPage = 0;
+	m_currentPageEntry = 0;
+
+#ifdef DEBUG_SpuCollisionTaskProcess
+	m_initialized = false;
+#endif
+
+#ifdef __CELLOS_LV2__
+// adding SPURS support.
+	gMidphaseSPU.startSPU();
+	
+#endif
+
+	//printf("sizeof vec_float4: %d\n", sizeof(vec_float4));
+	printf("sizeof SpuGatherAndProcessWorkUnitInput: %d\n", sizeof(SpuGatherAndProcessWorkUnitInput));
+
+}
+
+SpuCollisionTaskProcess::~SpuCollisionTaskProcess()
+{
+	if (m_workUnitTaskBuffers != 0)
+	{
+		free(m_workUnitTaskBuffers);
+		m_workUnitTaskBuffers = 0;
+	}
+	
+#ifdef __CELLOS_LV2__
+
+// Consolidating SPU code
+	gMidphaseSPU.stopSPU();
+	
+#endif
+}
+
+
+
+void
+SpuCollisionTaskProcess::initialize2()
+{
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("SpuCollisionTaskProcess::initialize()\n");
+#endif //DEBUG_SPU_TASK_SCHEDULING
+	if (!m_workUnitTaskBuffers)
+	{
+		m_workUnitTaskBuffers = (unsigned char *)memalign(128, MIDPHASE_WORKUNIT_TASK_SIZE*MIDPHASE_NUM_WORKUNIT_TASKS);
+	}
+
+	for (int i = 0; i < MIDPHASE_NUM_WORKUNIT_TASKS; i++)
+	{
+		m_taskBusy[i] = false;
+	}
+	m_numBusyTasks = 0;
+	m_currentTask = 0;
+	m_currentPage = 0;
+	m_currentPageEntry = 0;
+
+#ifdef DEBUG_SpuCollisionTaskProcess
+	m_initialized = true;
+	assert(MIDPHASE_NUM_WORKUNITS_PER_TASK*sizeof(SpuGatherAndProcessWorkUnitInput) <= MIDPHASE_WORKUNIT_TASK_SIZE);
+#endif
+}
+
+
+void SpuCollisionTaskProcess::issueTask2()
+{
+
+
+
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("SpuCollisionTaskProcess::issueTask (m_currentTask= %d\)n", m_currentTask);
+#endif //DEBUG_SPU_TASK_SCHEDULING
+
+	m_taskBusy[m_currentTask] = true;
+	m_numBusyTasks++;
+
+
+	SpuGatherAndProcessPairsTaskDesc& taskDesc = g_spuGatherTaskDesc[m_currentTask];
+	{
+		// send task description in event message
+		// no error checking here...
+		// but, currently, event queue can be no larger than NUM_WORKUNIT_TASKS.
+	
+		taskDesc.inPtr = reinterpret_cast<uint64_t>(MIDPHASE_TASK_PTR(m_currentTask));
+	
+		taskDesc.taskId = m_currentTask;
+		taskDesc.numPages = m_currentPage+1;
+		taskDesc.numOnLastPage = m_currentPageEntry;
+	}
+
+
+
+	gMidphaseSPU.sendRequest(CMD_GATHER_AND_PROCESS_PAIRLIST, (uint32_t) &taskDesc,m_currentTask);
+
+	// if all tasks busy, wait for spu event to clear the task.
+	
+
+	if (m_numBusyTasks >= MIDPHASE_NUM_WORKUNIT_TASKS)
+	{
+		unsigned int taskId;
+		unsigned int outputSize;
+
+		gMidphaseSPU.waitForResponse(&taskId, &outputSize);
+
+		//printf("PPU: after issue, received event: %u %d\n", taskId, outputSize);
+
+		//postProcess(taskId, outputSize);
+
+		m_taskBusy[taskId] = false;
+
+		m_numBusyTasks--;
+	}
+	
+}
+
+void SpuCollisionTaskProcess::addWorkToTask(void* pairArrayPtr,int startIndex,int endIndex)
+{
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("#");
+#endif //DEBUG_SPU_TASK_SCHEDULING
+	
+#ifdef DEBUG_SpuCollisionTaskProcess
+	assert(m_initialized);
+	assert(m_workUnitTaskBuffers);
+
+#endif
+
+	bool batch = true;
+
+	if (batch)
+	{
+		if (m_currentPageEntry == MIDPHASE_NUM_WORKUNITS_PER_PAGE)
+		{
+			if (m_currentPage == MIDPHASE_NUM_WORKUNIT_PAGES-1)
+			{
+				// task buffer is full, issue current task.
+				// if all task buffers busy, this waits until SPU is done.
+				issueTask2();
+
+				// find new task buffer
+				for (unsigned int i = 0; i < MIDPHASE_NUM_WORKUNIT_TASKS; i++)
+				{
+					if (!m_taskBusy[i])
+					{
+						m_currentTask = i;
+						//init the task data
+
+						break;
+					}
+				}
+
+				m_currentPage = 0;
+			}
+			else
+			{
+				m_currentPage++;
+			}
+
+			m_currentPageEntry = 0;
+		}
+	}
+
+	{
+
+
+
+		SpuGatherAndProcessWorkUnitInput &wuInput = 
+			*(reinterpret_cast<SpuGatherAndProcessWorkUnitInput*>
+			(MIDPHASE_ENTRY_PTR(m_currentTask, m_currentPage, m_currentPageEntry)));
+		
+		wuInput.m_pairArrayPtr = reinterpret_cast<uint64_t>(pairArrayPtr);
+		wuInput.m_startIndex = startIndex;
+		wuInput.m_endIndex = endIndex;
+
+		
+	
+		m_currentPageEntry++;
+
+		if (!batch)
+		{
+			issueTask2();
+
+			// find new task buffer
+			for (unsigned int i = 0; i < MIDPHASE_NUM_WORKUNIT_TASKS; i++)
+			{
+				if (!m_taskBusy[i])
+				{
+					m_currentTask = i;
+					//init the task data
+
+					break;
+				}
+			}
+
+			m_currentPage = 0;
+			m_currentPageEntry =0;
+		}
+	}
+}
+
+
+void 
+SpuCollisionTaskProcess::flush2()
+{
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("\nSpuCollisionTaskProcess::flush()\n");
+#endif //DEBUG_SPU_TASK_SCHEDULING
+	
+	// if there's a partially filled task buffer, submit that task
+	if (m_currentPage > 0 || m_currentPageEntry > 0)
+	{
+		issueTask2();
+	}
+
+
+	// all tasks are issued, wait for all tasks to be complete
+	while(m_numBusyTasks > 0)
+	{
+// Consolidating SPU code
+	  unsigned int taskId;
+	  unsigned int outputSize;
+	  
+	  {
+			
+		// SPURS support.
+		  gMidphaseSPU.waitForResponse(&taskId, &outputSize);
+	  }
+
+		//printf("PPU: flushing, received event: %u %d\n", taskId, outputSize);
+
+		//postProcess(taskId, outputSize);
+
+		m_taskBusy[taskId] = false;
+
+		m_numBusyTasks--;
+	}
+
+
+}
diff --git a/Extras/BulletMultiThreaded/SpuCollisionTaskProcess.h b/Extras/BulletMultiThreaded/SpuCollisionTaskProcess.h
new file mode 100644
index 000000000..e1fe040c2
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuCollisionTaskProcess.h
@@ -0,0 +1,131 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef SPU_COLLISION_TASK_PROCESS_H
+#define SPU_COLLISION_TASK_PROCESS_H
+
+#include <assert.h>
+
+#include <LinearMath/btScalar.h>
+
+#include "PlatformDefinitions.h"
+
+
+#define DEBUG_SpuCollisionTaskProcess 1
+
+//#define MIDPHASE_NUM_WORKUNIT_TASKS 4 //attempt for multi-core
+#define MIDPHASE_NUM_WORKUNIT_TASKS 12 //attempt for multi-core
+//#define MIDPHASE_NUM_WORKUNIT_TASKS 6 //for SPUs
+
+#define CMD_GATHER_AND_PROCESS_PAIRLIST	1
+
+class btCollisionObject;
+class btPersistentManifold;
+class btDispatcher;
+
+
+///Task Description for SPU collision detection
+struct SpuGatherAndProcessPairsTaskDesc
+{
+	uint64_t	inPtr;//m_pairArrayPtr;
+	//mutex variable
+	uint32_t	m_someMutexVariableInMainMemory;
+
+	uint64_t	m_dispatcher;
+
+	uint32_t	numOnLastPage;
+
+	uint16_t numPages;
+	uint16_t taskId;
+
+	struct	CollisionTask_LocalStoreMemory*	m_lsMemory;
+}
+#ifdef __CELLOS_LV2__
+__attribute__ ((aligned (16)))
+#endif
+;
+
+
+///MidphaseWorkUnitInput stores individual primitive versus mesh collision detection input, to be processed by the SPU.
+struct SpuGatherAndProcessWorkUnitInput
+{
+	uint64_t m_pairArrayPtr;
+	int		m_startIndex;
+	int		m_endIndex;
+};
+
+
+
+
+/// SpuCollisionTaskProcess handles SPU processing of collision pairs.
+/// Maintains a set of task buffers.
+/// When the task is full, the task is issued for SPUs to process.  Contact output goes into btPersistentManifold
+/// associated with each task.
+/// When PPU issues a task, it will look for completed task buffers
+/// PPU will do postprocessing, dependent on workunit output (not likely)
+class SpuCollisionTaskProcess
+{
+
+  unsigned char  *m_workUnitTaskBuffers;
+
+
+	// track task buffers that are being used, and total busy tasks
+	bool           m_taskBusy[MIDPHASE_NUM_WORKUNIT_TASKS];
+	unsigned int   m_numBusyTasks;
+
+	// the current task and the current entry to insert a new work unit
+	unsigned int   m_currentTask;
+	unsigned int   m_currentPage;
+	unsigned int   m_currentPageEntry;
+
+#ifdef DEBUG_SpuCollisionTaskProcess
+	bool m_initialized;
+#endif
+	void issueTask2();
+	//void postProcess(unsigned int taskId, int outputSize);
+
+public:
+	SpuCollisionTaskProcess();
+	
+	~SpuCollisionTaskProcess();
+	
+	///call initialize in the beginning of the frame, before addCollisionPairToTask
+	void initialize2();
+
+	///batch up additional work to a current task for SPU processing. When batch is full, it issues the task.
+	void addWorkToTask(void* pairArrayPtr,int startIndex,int endIndex);
+
+	///call flush to submit potential outstanding work to SPUs and wait for all involved SPUs to be finished
+	void flush2();
+};
+
+
+
+#define MIDPHASE_TASK_PTR(task) (&m_workUnitTaskBuffers[0] + MIDPHASE_WORKUNIT_TASK_SIZE*task)
+#define MIDPHASE_ENTRY_PTR(task,page,entry) (MIDPHASE_TASK_PTR(task) + MIDPHASE_WORKUNIT_PAGE_SIZE*page + sizeof(SpuGatherAndProcessWorkUnitInput)*entry)
+#define MIDPHASE_OUTPUT_PTR(task) (&m_contactOutputBuffers[0] + MIDPHASE_MAX_CONTACT_BUFFER_SIZE*task)
+#define MIDPHASE_TREENODES_PTR(task) (&m_complexShapeBuffers[0] + MIDPHASE_COMPLEX_SHAPE_BUFFER_SIZE*task)
+
+
+#define MIDPHASE_WORKUNIT_PAGE_SIZE (16)
+
+#define MIDPHASE_NUM_WORKUNIT_PAGES 1
+#define MIDPHASE_WORKUNIT_TASK_SIZE (MIDPHASE_WORKUNIT_PAGE_SIZE*MIDPHASE_NUM_WORKUNIT_PAGES)
+#define MIDPHASE_NUM_WORKUNITS_PER_PAGE (MIDPHASE_WORKUNIT_PAGE_SIZE / sizeof(SpuGatherAndProcessWorkUnitInput))
+#define MIDPHASE_NUM_WORKUNITS_PER_TASK (MIDPHASE_NUM_WORKUNITS_PER_PAGE*MIDPHASE_NUM_WORKUNIT_PAGES)
+
+
+#endif // SPU_COLLISION_TASK_PROCESS_H
+
diff --git a/Extras/BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.cpp b/Extras/BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.cpp
new file mode 100644
index 000000000..6bf85dbd2
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.cpp
@@ -0,0 +1,58 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "SpuContactManifoldCollisionAlgorithm.h"
+#include "BulletCollision/CollisionDispatch/btCollisionDispatcher.h"
+#include "BulletCollision/CollisionDispatch/btCollisionObject.h"
+#include "BulletCollision/CollisionShapes/btCollisionShape.h"
+
+
+SpuContactManifoldCollisionAlgorithm::SpuContactManifoldCollisionAlgorithm()
+:m_manifoldPtr(0)
+{
+	
+}
+
+
+void SpuContactManifoldCollisionAlgorithm::processCollision (btCollisionObject* body0,btCollisionObject* body1,const btDispatcherInfo& dispatchInfo,btManifoldResult* resultOut)
+{
+	btAssert(0);
+}
+
+float SpuContactManifoldCollisionAlgorithm::calculateTimeOfImpact(btCollisionObject* body0,btCollisionObject* body1,const btDispatcherInfo& dispatchInfo,btManifoldResult* resultOut)
+{
+	btAssert(0);
+	return 1.f;
+}
+
+#ifndef __SPU__
+SpuContactManifoldCollisionAlgorithm::SpuContactManifoldCollisionAlgorithm(const btCollisionAlgorithmConstructionInfo& ci,btCollisionObject* body0,btCollisionObject* body1)
+:btCollisionAlgorithm(ci)
+{
+	m_manifoldPtr = m_dispatcher->getNewManifold(body0,body1);
+	m_shapeType0 = body0->getCollisionShape()->getShapeType();
+	m_shapeType1 = body1->getCollisionShape()->getShapeType();
+	m_collisionMargin0 = body0->getCollisionShape()->getMargin();
+	m_collisionMargin1 = body1->getCollisionShape()->getMargin();
+
+}
+#endif //__SPU__
+
+
+SpuContactManifoldCollisionAlgorithm::~SpuContactManifoldCollisionAlgorithm()
+{
+	if (m_manifoldPtr)
+			m_dispatcher->releaseManifold(m_manifoldPtr);
+}
\ No newline at end of file
diff --git a/Extras/BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.h b/Extras/BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.h
new file mode 100644
index 000000000..0f0c933cf
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuContactManifoldCollisionAlgorithm.h
@@ -0,0 +1,80 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef SPU_CONTACTMANIFOLD_COLLISION_ALGORITHM_H
+#define SPU_CONTACTMANIFOLD_COLLISION_ALGORITHM_H
+
+#include "BulletCollision/BroadphaseCollision/btCollisionAlgorithm.h"
+#include "BulletCollision/BroadphaseCollision/btBroadphaseProxy.h"
+#include "BulletCollision/CollisionDispatch/btCollisionCreateFunc.h"
+class btPersistentManifold;
+
+/// SpuContactManifoldCollisionAlgorithm  provides contact manifold and should be processed on SPU.
+ATTRIBUTE_ALIGNED16(class) SpuContactManifoldCollisionAlgorithm : public btCollisionAlgorithm
+{
+
+	btPersistentManifold*	m_manifoldPtr;
+	int		m_shapeType0;
+	int		m_shapeType1;
+	float	m_collisionMargin0;
+	float	m_collisionMargin1;
+
+	
+public:
+	
+	virtual void processCollision (btCollisionObject* body0,btCollisionObject* body1,const btDispatcherInfo& dispatchInfo,btManifoldResult* resultOut);
+
+	virtual float calculateTimeOfImpact(btCollisionObject* body0,btCollisionObject* body1,const btDispatcherInfo& dispatchInfo,btManifoldResult* resultOut);
+
+	SpuContactManifoldCollisionAlgorithm();
+
+	SpuContactManifoldCollisionAlgorithm(const btCollisionAlgorithmConstructionInfo& ci,btCollisionObject* body0,btCollisionObject* body1);
+
+	virtual ~SpuContactManifoldCollisionAlgorithm();
+
+	btPersistentManifold*	getContactManifoldPtr()
+	{
+		return m_manifoldPtr;
+	}
+
+	int		getShapeType0() const
+	{
+		return m_shapeType0;
+	}
+
+	int		getShapeType1() const
+	{
+		return m_shapeType1;
+	}
+	float	getCollisionMargin0() const
+	{
+		return m_collisionMargin0;
+	}
+	float	getCollisionMargin1() const
+	{
+		return m_collisionMargin1;
+	}
+
+	struct CreateFunc :public 	btCollisionAlgorithmCreateFunc
+	{
+		virtual	btCollisionAlgorithm* CreateCollisionAlgorithm(btCollisionAlgorithmConstructionInfo& ci, btCollisionObject* body0,btCollisionObject* body1)
+		{
+			return new SpuContactManifoldCollisionAlgorithm(ci,body0,body1);
+		}
+	};
+
+};
+
+#endif //SPU_CONTACTMANIFOLD_COLLISION_ALGORITHM_H
diff --git a/Extras/BulletMultiThreaded/SpuDoubleBuffer.h b/Extras/BulletMultiThreaded/SpuDoubleBuffer.h
new file mode 100644
index 000000000..92f465cd4
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuDoubleBuffer.h
@@ -0,0 +1,105 @@
+#ifndef DOUBLE_BUFFER_H
+#define DOUBLE_BUFFER_H
+
+#ifdef __CELLOS_LV2__
+
+#include <cell/dma.h>
+#include <stdint.h>
+
+#else
+#include "SpuFakeDma.h"
+#endif //
+
+
+///DoubleBuffer
+template<class T, int size>
+class DoubleBuffer
+{
+#ifdef __CELLOS_LV2__
+	T m_buffer0[size] __attribute__ ((aligned (128)));
+	T m_buffer1[size] __attribute__ ((aligned (128)));
+#else
+	T m_buffer0[size];
+	T m_buffer1[size];
+#endif
+	
+	T *m_frontBuffer;
+	T *m_backBuffer;
+
+	unsigned int m_dmaTag;
+	bool m_dmaPending;
+public:
+	bool	isPending() const { return m_dmaPending;}
+	DoubleBuffer();
+
+	// dma get and put commands
+	void backBufferDmaGet(uint64_t ea, unsigned int numBytes, unsigned int tag);
+	void backBufferDmaPut(uint64_t ea, unsigned int numBytes, unsigned int tag);
+
+	// gets pointer to a buffer
+	T *getFront();
+	T *getBack();
+
+	// if back buffer dma was started, wait for it to complete
+	// then move back to front and vice versa
+	T *swapBuffers();
+};
+
+template<class T, int size>
+DoubleBuffer<T,size>::DoubleBuffer()
+{
+	m_dmaPending = false;
+	m_frontBuffer = &m_buffer0[0];
+	m_backBuffer = &m_buffer1[0];
+}
+
+template<class T, int size>
+void
+DoubleBuffer<T,size>::backBufferDmaGet(uint64_t ea, unsigned int numBytes, unsigned int tag)
+{
+	m_dmaPending = true;
+	m_dmaTag = tag;
+	cellDmaLargeGet(m_backBuffer, ea, numBytes, tag, 0, 0);
+}
+
+template<class T, int size>
+void
+DoubleBuffer<T,size>::backBufferDmaPut(uint64_t ea, unsigned int numBytes, unsigned int tag)
+{
+	m_dmaPending = true;
+	m_dmaTag = tag;
+	cellDmaLargePut(m_backBuffer, ea, numBytes, tag, 0, 0);
+}
+
+template<class T, int size>
+T *
+DoubleBuffer<T,size>::getFront()
+{
+	return m_frontBuffer;
+}
+
+template<class T, int size>
+T *
+DoubleBuffer<T,size>::getBack()
+{
+	return m_backBuffer;
+}
+
+template<class T, int size>
+T *
+DoubleBuffer<T,size>::swapBuffers()
+{
+	if (m_dmaPending)
+	{
+		cellDmaWaitTagStatusAll(1<<m_dmaTag);
+		m_dmaPending = false;
+	}
+
+	T *tmp = m_backBuffer;
+	m_backBuffer = m_frontBuffer;
+	m_frontBuffer = tmp;
+
+	return m_frontBuffer;
+}
+
+#endif
diff --git a/Extras/BulletMultiThreaded/SpuFakeDma.cpp b/Extras/BulletMultiThreaded/SpuFakeDma.cpp
new file mode 100644
index 000000000..2b9456874
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuFakeDma.cpp
@@ -0,0 +1,30 @@
+
+#include "SpuFakeDma.h"
+
+int	cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
+{
+	void* targetMainMem = (void*)ea;
+	memcpy(ls,targetMainMem,size);
+	return 0;
+}
+
+int	cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
+{
+	void* targetMainMem = (void*)ea;
+	memcpy(ls,targetMainMem,size);
+	return 0;
+}
+
+int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid)
+{
+	void* targetMainMem = (void*)ea;
+	memcpy(targetMainMem,ls,size);
+	return 0;
+}
+
+
+
+void	cellDmaWaitTagStatusAll(int ignore)
+{
+
+}
diff --git a/Extras/BulletMultiThreaded/SpuFakeDma.h b/Extras/BulletMultiThreaded/SpuFakeDma.h
new file mode 100644
index 000000000..5bc73ce37
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuFakeDma.h
@@ -0,0 +1,21 @@
+
+#ifndef FAKE_DMA_H
+#define FAKE_DMA_H
+
+
+#include "PlatformDefinitions.h"
+
+#include <LinearMath/btScalar.h> //for definition of uint64_t,uint32_t
+#define DMA_TAG(a) (a)
+#define DMA_MASK(a) (a)
+
+/// cellDmaLargeGet Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
+int	cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
+int	cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
+/// cellDmaLargePut Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
+int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid);
+/// cellDmaWaitTagStatusAll Win32 replacements for Cell DMA to allow simulating most of the SPU code (just memcpy)
+void	cellDmaWaitTagStatusAll(int ignore);
+
+
+#endif //FAKE_DMA_H
\ No newline at end of file
diff --git a/Extras/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp b/Extras/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp
new file mode 100644
index 000000000..c3b6e3845
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuGatheringCollisionDispatcher.cpp
@@ -0,0 +1,210 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "SpuGatheringCollisionDispatcher.h"
+#include "SpuCollisionTaskProcess.h"
+
+
+#include "BulletCollision/BroadphaseCollision/btOverlappingPairCache.h"
+#include "BulletCollision/CollisionDispatch/btEmptyCollisionAlgorithm.h"
+#include "SpuContactManifoldCollisionAlgorithm.h"
+#include "BulletCollision/CollisionDispatch/btCollisionObject.h"
+#include "BulletCollision/CollisionShapes/btCollisionShape.h"
+
+
+
+
+SpuGatheringCollisionDispatcher::SpuGatheringCollisionDispatcher()
+:m_spuCollisionTaskProcess(0)
+{
+	
+}
+
+
+bool	SpuGatheringCollisionDispatcher::supportsDispatchPairOnSpu(int proxyType0,int proxyType1)
+{
+	bool supported0 = (
+		(proxyType0 == BOX_SHAPE_PROXYTYPE) ||
+		(proxyType0 == TRIANGLE_SHAPE_PROXYTYPE) ||
+		(proxyType0 == SPHERE_SHAPE_PROXYTYPE) ||
+		(proxyType0 == CAPSULE_SHAPE_PROXYTYPE) ||
+		(proxyType0 == CYLINDER_SHAPE_PROXYTYPE) ||
+//		(proxyType0 == CONE_SHAPE_PROXYTYPE) ||
+		(proxyType0 == TRIANGLE_MESH_SHAPE_PROXYTYPE) ||
+		(proxyType0 == CONVEX_HULL_SHAPE_PROXYTYPE)
+		);
+
+	bool supported1 = (
+		(proxyType1 == BOX_SHAPE_PROXYTYPE) ||
+		(proxyType1 == TRIANGLE_SHAPE_PROXYTYPE) ||
+		(proxyType1 == SPHERE_SHAPE_PROXYTYPE) ||
+		(proxyType1 == CAPSULE_SHAPE_PROXYTYPE) ||
+		(proxyType1 == CYLINDER_SHAPE_PROXYTYPE) ||
+//		(proxyType1 == CONE_SHAPE_PROXYTYPE) ||
+		(proxyType1 == TRIANGLE_MESH_SHAPE_PROXYTYPE) ||
+		(proxyType1 == CONVEX_HULL_SHAPE_PROXYTYPE)
+		);
+
+	return supported0 && supported1;
+}
+
+
+
+SpuGatheringCollisionDispatcher::~SpuGatheringCollisionDispatcher()
+{
+	if (m_spuCollisionTaskProcess)
+		delete m_spuCollisionTaskProcess;
+	
+}
+
+#include "stdio.h"
+
+
+
+///interface for iterating all overlapping collision pairs, no matter how those pairs are stored (array, set, map etc)
+///this is useful for the collision dispatcher.
+class btSpuCollisionPairCallback : public btOverlapCallback
+{
+	btDispatcherInfo& m_dispatchInfo;
+	SpuGatheringCollisionDispatcher*	m_dispatcher;
+
+public:
+
+	btSpuCollisionPairCallback(btDispatcherInfo& dispatchInfo,SpuGatheringCollisionDispatcher*	dispatcher)
+	:m_dispatchInfo(dispatchInfo),
+	m_dispatcher(dispatcher)
+	{
+	}
+
+	virtual bool	processOverlap(btBroadphasePair& collisionPair)
+	{
+
+
+		//PPU version
+		//(*m_dispatcher->getNearCallback())(collisionPair,*m_dispatcher,m_dispatchInfo);
+
+		//only support discrete collision detection for now, we could fallback on PPU/unoptimized version for TOI/CCD
+		btAssert(m_dispatchInfo.m_dispatchFunc == btDispatcherInfo::DISPATCH_DISCRETE);
+
+		//by default, Bullet will use this near callback
+		{
+			///userInfo is used to determine if the SPU has to handle this case or not (skip PPU tasks)
+			if (!collisionPair.m_userInfo)
+			{
+				collisionPair.m_userInfo = (void*) 1;
+			}
+			if (!collisionPair.m_algorithm)
+			{
+				btCollisionObject* colObj0 = (btCollisionObject*)collisionPair.m_pProxy0->m_clientObject;
+				btCollisionObject* colObj1 = (btCollisionObject*)collisionPair.m_pProxy1->m_clientObject;
+
+				btCollisionAlgorithmConstructionInfo ci;
+				ci.m_dispatcher = m_dispatcher;
+				ci.m_manifold = 0;
+
+				if (m_dispatcher->needsCollision(colObj0,colObj1))
+				{
+					int	proxyType0 = colObj0->getCollisionShape()->getShapeType();
+					int	proxyType1 = colObj1->getCollisionShape()->getShapeType();
+					if (m_dispatcher->supportsDispatchPairOnSpu(proxyType0,proxyType1))
+					{
+						collisionPair.m_algorithm = new SpuContactManifoldCollisionAlgorithm(ci,colObj0,colObj1);
+						collisionPair.m_userInfo = (void*) 2;
+					} else
+					{
+						collisionPair.m_algorithm = m_dispatcher->findAlgorithm(colObj0,colObj1);
+						collisionPair.m_userInfo = (void*)3;
+					}
+				} else
+				{
+					//create an empty algorithm
+					collisionPair.m_algorithm = new btEmptyAlgorithm(ci);
+				}
+			}
+		}
+		return false;
+	}
+};
+
+void	SpuGatheringCollisionDispatcher::dispatchAllCollisionPairs(btOverlappingPairCache* pairCache,btDispatcherInfo& dispatchInfo)
+{
+	//cellPerfInsertCBEpmBookmark(1);
+
+	if (dispatchInfo.m_enableSPU)
+	{
+		if (!m_spuCollisionTaskProcess)
+			m_spuCollisionTaskProcess = new SpuCollisionTaskProcess();
+	
+		m_spuCollisionTaskProcess->initialize2();
+	
+		///modified version of btCollisionDispatcher::dispatchAllCollisionPairs:
+		{
+			btSpuCollisionPairCallback	collisionCallback(dispatchInfo,this);
+
+			pairCache->processAllOverlappingPairs(&collisionCallback);
+		}
+
+		//send one big batch
+		int numTotalPairs = pairCache->getNumOverlappingPairs();
+		btBroadphasePair* pairPtr = pairCache->getOverlappingPairArrayPtr();
+		int i;
+		for (i=0;i<numTotalPairs;)
+		{
+			//Performance Hint: tweak this number during benchmarking
+			static const int pairRange = SPU_BATCHSIZE_BROADPHASE_PAIRS;
+			int endIndex = (i+pairRange) < numTotalPairs ? i+pairRange : numTotalPairs;
+			m_spuCollisionTaskProcess->addWorkToTask(pairPtr,i,endIndex);
+			i = endIndex;
+		}
+
+		//handle PPU fallback pairs
+		for (i=0;i<numTotalPairs;i++)
+		{
+			btBroadphasePair& collisionPair = pairPtr[i];
+			if (collisionPair.m_userInfo == (void*)3)
+			{
+				if (collisionPair.m_algorithm)
+				{
+					btCollisionObject* colObj0 = (btCollisionObject*)collisionPair.m_pProxy0->m_clientObject;
+					btCollisionObject* colObj1 = (btCollisionObject*)collisionPair.m_pProxy1->m_clientObject;
+					btManifoldResult contactPointResult(colObj0,colObj1);
+					
+					if (dispatchInfo.m_dispatchFunc == 		btDispatcherInfo::DISPATCH_DISCRETE)
+					{
+						//discrete collision detection query
+						collisionPair.m_algorithm->processCollision(colObj0,colObj1,dispatchInfo,&contactPointResult);
+					} else
+					{
+						//continuous collision detection query, time of impact (toi)
+						btScalar toi = collisionPair.m_algorithm->calculateTimeOfImpact(colObj0,colObj1,dispatchInfo,&contactPointResult);
+						if (dispatchInfo.m_timeOfImpact > toi)
+							dispatchInfo.m_timeOfImpact = toi;
+
+					}
+				}
+			}
+		}
+
+		//make sure all SPU work is done
+		m_spuCollisionTaskProcess->flush2();
+
+	} else
+	{
+		///PPU fallback
+		///!Need to make sure to clear all 'algorithms' when switching between SPU and PPU
+		btCollisionDispatcher::dispatchAllCollisionPairs(pairCache,dispatchInfo);
+	}
+}
+
diff --git a/Extras/BulletMultiThreaded/SpuGatheringCollisionDispatcher.h b/Extras/BulletMultiThreaded/SpuGatheringCollisionDispatcher.h
new file mode 100644
index 000000000..88e3191f8
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuGatheringCollisionDispatcher.h
@@ -0,0 +1,58 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+#ifndef SPU_GATHERING_COLLISION__DISPATCHER_H
+#define SPU_GATHERING_COLLISION__DISPATCHER_H
+
+#include "BulletCollision/CollisionDispatch/btCollisionDispatcher.h"
+
+
+///Tuning value to optimized SPU utilization 
+///Too small value means Task overhead is large compared to computation (too fine granularity)
+///Too big value might render some SPUs are idle, while a few other SPUs are doing all work.
+//#define SPU_BATCHSIZE_BROADPHASE_PAIRS 16
+#define SPU_BATCHSIZE_BROADPHASE_PAIRS 256
+
+
+class SpuCollisionTaskProcess;
+
+///SpuGatheringCollisionDispatcher can use SPU to gather and calculate collision detection
+///Time of Impact, Closest Points and Penetration Depth.
+class SpuGatheringCollisionDispatcher : public btCollisionDispatcher
+{
+	
+	SpuCollisionTaskProcess*	m_spuCollisionTaskProcess;
+	
+
+public:
+
+	//can be used by SPU collision algorithms	
+	SpuCollisionTaskProcess*	getSpuCollisionTaskProcess()
+	{
+			return m_spuCollisionTaskProcess;
+	}
+	
+	SpuGatheringCollisionDispatcher ();
+
+	virtual ~SpuGatheringCollisionDispatcher();
+
+	bool	supportsDispatchPairOnSpu(int proxyType0,int proxyType1);
+
+	virtual void	dispatchAllCollisionPairs(btOverlappingPairCache* pairCache,btDispatcherInfo& dispatchInfo);
+
+};
+
+
+
+#endif //SPU_GATHERING_COLLISION__DISPATCHER_H
diff --git a/Extras/BulletMultiThreaded/SpuIntegrationTask/readme.txt b/Extras/BulletMultiThreaded/SpuIntegrationTask/readme.txt
new file mode 100644
index 000000000..f3e907347
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuIntegrationTask/readme.txt
@@ -0,0 +1 @@
+Empty placeholder for future Libspe2 SPU task
diff --git a/Extras/BulletMultiThreaded/SpuLibspe2Support.cpp b/Extras/BulletMultiThreaded/SpuLibspe2Support.cpp
new file mode 100644
index 000000000..f22428ca2
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuLibspe2Support.cpp
@@ -0,0 +1,224 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+#ifdef USE_LIBSPE2
+
+#include "SpuLibspe2Support.h"
+
+#include "SpuCollisionTaskProcess.h"
+#include "SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h"
+
+#include <Windows.h>
+
+///SpuLibspe2Support helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
+///Setup and initialize SPU/CELL/Libspe2
+SpuLibspe2Support::SpuLibspe2Support(SpuLibspe2ElfId_t elfId,int numThreads)
+{
+	startSPUs(numThreads);
+}
+
+///cleanup/shutdown Libspe2
+SpuLibspe2Support::~SpuLibspe2Support()
+{
+	stopSPUs();
+}
+
+
+
+#include <stdio.h>
+
+DWORD WINAPI Thread_no_1( LPVOID lpParam ) 
+{
+
+	btSpuStatus* status = (btSpuStatus*)lpParam;
+
+	
+	while (1)
+	{
+		WaitForSingleObject(status->m_eventStartHandle,INFINITE);
+		btAssert(status->m_status);
+
+		SpuGatherAndProcessPairsTaskDesc* taskDesc = status->m_taskDesc;
+
+		if (taskDesc)
+		{
+			processCollisionTask(*taskDesc);
+			SetEvent(status->m_eventCompletetHandle);
+		} else
+		{
+			//exit Thread
+			break;
+		}
+		
+	}
+
+	return 0;
+
+}
+
+///send messages to SPUs
+void SpuLibspe2Support::sendRequest(uint32_t uiCommand, uint32_t uiArgument0, uint32_t uiArgument1)
+{
+	///	gMidphaseSPU.sendRequest(CMD_GATHER_AND_PROCESS_PAIRLIST, (uint32_t) &taskDesc);
+	
+	///we should spawn an SPU task here, and in 'waitForResponse' it should wait for response of the (one of) the first tasks that finished
+	
+
+
+	switch (uiCommand)
+	{
+	case 	CMD_GATHER_AND_PROCESS_PAIRLIST:
+		{
+
+			SpuGatherAndProcessPairsTaskDesc* taskDesc = (SpuGatherAndProcessPairsTaskDesc*) uiArgument0 ;
+
+//#define SINGLE_THREADED 1
+#ifdef SINGLE_THREADED
+
+			btSpuStatus&	spuStatus = m_activeSpuStatus[0];
+			taskDesc->m_lsMemory = (CollisionTask_LocalStoreMemory*)spuStatus.m_lsMemory;
+			processCollisionTask(*taskDesc);
+			HANDLE handle =0;
+#else
+
+			btAssert(taskDesc->taskId>=0);
+			btAssert(taskDesc->taskId<m_activeSpuStatus.size());
+
+			btSpuStatus&	spuStatus = m_activeSpuStatus[taskDesc->taskId];
+
+			spuStatus.m_commandId = uiCommand;
+			spuStatus.m_status = 1;
+			spuStatus.m_taskDesc = taskDesc;
+			taskDesc->m_lsMemory = (CollisionTask_LocalStoreMemory*)spuStatus.m_lsMemory;
+			///fire event to start new task
+			SetEvent(spuStatus.m_eventStartHandle);
+
+#endif //CollisionTask_LocalStoreMemory
+
+			
+
+			break;
+		}
+	default:
+		{
+			///not implemented
+			btAssert(0);
+		}
+
+	};
+
+
+}
+
+///check for messages from SPUs
+void SpuLibspe2Support::waitForResponse(unsigned int *puiArgument0, unsigned int *puiArgument1)
+{
+	///We should wait for (one of) the first tasks to finish (or other SPU messages), and report its response
+	
+	///A possible response can be 'yes, SPU handled it', or 'no, please do a PPU fallback'
+
+
+	btAssert(m_activeSpuStatus.size());
+
+	int last = -1;
+	
+	//find an active spu/thread
+	for (int i=0;i<m_activeSpuStatus.size();i++)
+	{
+		if (m_activeSpuStatus[i].m_status)
+		{
+			last = i;
+			break;
+		}
+	}
+
+
+#ifndef SINGLE_THREADED
+	btAssert(spuStatus.m_threadHandle);
+	btAssert(spuStatus.m_eventCompletetHandle);
+	btSpuStatus& spuStatus = m_activeSpuStatus[last];
+	WaitForSingleObject(spuStatus.m_eventCompletetHandle, INFINITE);
+	spuStatus.m_status = 0;
+
+	///need to find an active spu
+	btAssert(last>=0);
+
+#else
+	last=0;
+	btSpuStatus& spuStatus = m_activeSpuStatus[last];
+#endif //SINGLE_THREADED
+
+	
+
+	*puiArgument0 = spuStatus.m_taskId;
+	*puiArgument1 = spuStatus.m_status;
+
+
+}
+
+
+///start the spus group (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+void SpuLibspe2Support::startSPUs(int numThreads)
+{
+
+	m_activeSpuStatus.resize(numThreads);
+
+	for (int i=0;i<numThreads;i++)
+	{
+		printf("starting thread %d\n",i);
+
+		btSpuStatus&	spuStatus = m_activeSpuStatus[i];
+
+		LPSECURITY_ATTRIBUTES lpThreadAttributes=NULL;
+		SIZE_T dwStackSize=65535;
+		LPTHREAD_START_ROUTINE lpStartAddress=&Thread_no_1;
+		LPVOID lpParameter=&spuStatus;
+		DWORD dwCreationFlags=0;
+		LPDWORD lpThreadId=0;
+
+		spuStatus.m_taskDesc = 0;
+
+		sprintf(spuStatus.m_eventStartHandleName,"eventStart%d",i);
+		spuStatus.m_eventStartHandle = CreateEvent(0,false,false,spuStatus.m_eventStartHandleName);
+
+		sprintf(spuStatus.m_eventCompletetHandleName,"eventComplete%d",i);
+		spuStatus.m_eventCompletetHandle = CreateEvent(0,false,false,spuStatus.m_eventCompletetHandleName);
+
+
+		HANDLE handle = CreateThread(lpThreadAttributes,dwStackSize,lpStartAddress,lpParameter,	dwCreationFlags,lpThreadId);
+		SetThreadPriority(handle,THREAD_PRIORITY_TIME_CRITICAL);
+
+		spuStatus.m_taskId = i;
+		spuStatus.m_commandId = 0;
+		spuStatus.m_status = 0;
+		spuStatus.m_threadHandle = handle;
+		spuStatus.m_lsMemory = createLocalStoreMemory();
+
+		printf("started thread %d with threadHandle %d\n",i,handle);
+		
+	}
+
+}
+
+///tell the task scheduler we are done with the SPU tasks
+void SpuLibspe2Support::stopSPUs()
+{
+//	m_activeSpuStatus.pop_back();
+//	WaitForSingleObject(spuStatus.bla, INFINITE);
+//	CloseHandle(spuStatus.m_threadHandle);
+	
+}
+
+#endif// USE_LIBSPE2
+
diff --git a/Extras/BulletMultiThreaded/SpuLibspe2Support.h b/Extras/BulletMultiThreaded/SpuLibspe2Support.h
new file mode 100644
index 000000000..fd8af3f33
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuLibspe2Support.h
@@ -0,0 +1,102 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+///USE_LIBSPE2: this define should be in the build system, or in <LinearMath/btScalar.h>
+//#define USE_LIBSPE2 1
+#ifdef USE_LIBSPE2
+
+#ifndef SPU_LIBSPE2_SUPPORT_H
+#define SPU_LIBSPE2_SUPPORT_H
+
+#include "LinearMath/btAlignedObjectArray.h"
+
+
+/**
+ * Note:
+ * The order of elements in this enum are important, that's why each one is explicitly
+ * given a value.  They will correspond to the .elf names/addresses that will be
+ * loaded into Libspe2.
+ * Mixing up these values will cause the wrong code to execute, for instance, the
+ * solver may be asked to do a collision detection job.
+ */
+typedef enum {
+	SPU_ELF_COLLISION_DETECTION=0,
+	SPU_ELF_SAMPLE,
+//SPU_ELF_INTEGRATION,
+//SPU_ELF_SOLVER,
+	SPU_ELF_LAST,
+} SpuLibspe2ElfId_t;
+
+#ifdef WIN32
+#include <malloc.h>
+#define memalign(alignment, size) malloc(size);
+#else
+#include <stdlib.h>
+#endif // WIN32
+
+
+
+#include <LinearMath/btScalar.h> //for uint32_t etc.
+
+///placeholder, until libspe2 support is there
+struct	btSpuStatus
+{
+	uint32_t m_taskId;
+	uint32_t	m_commandId;
+	uint32_t m_status;
+
+	struct SpuGatherAndProcessPairsTaskDesc* m_taskDesc;
+
+	void*	m_threadHandle;
+	void*	m_lsMemory;
+
+	void*	m_eventStartHandle;
+	char	m_eventStartHandleName[32];
+
+	void*	m_eventCompletetHandle;
+	char	m_eventCompletetHandleName[32];
+	
+
+};
+
+///SpuLibspe2Support helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
+class SpuLibspe2Support {
+
+	btAlignedObjectArray<btSpuStatus>	m_activeSpuStatus;
+
+public:
+	///Setup and initialize SPU/CELL/Libspe2
+	SpuLibspe2Support(SpuLibspe2ElfId_t elfId,int numThreads);
+
+///cleanup/shutdown Libspe2
+	~SpuLibspe2Support();
+
+///send messages to SPUs
+	void sendRequest(uint32_t uiCommand, uint32_t uiArgument0, uint32_t uiArgument1=0);
+
+///check for messages from SPUs
+	void waitForResponse(unsigned int *puiArgument0, unsigned int *puiArgument1);
+
+///start the spus (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+	void startSPUs(int numThreads);
+
+///tell the task scheduler we are done with the SPU tasks
+	void stopSPUs();
+
+};
+
+#endif //SPU_LIBSPE2_SUPPORT_H
+
+#endif //USE_LIBSPE2
\ No newline at end of file
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuContactResult.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuContactResult.cpp
new file mode 100644
index 000000000..c00cca066
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuContactResult.cpp
@@ -0,0 +1,162 @@
+/* [SCE CONFIDENTIAL DOCUMENT]
+ * PLAYSTATION(R)3 SPU Optimized Bullet Physics Library (http://bulletphysics.com)
+ *                Copyright (C) 2007 Sony Computer Entertainment Inc.
+ *                                                All Rights Reserved.
+ */
+
+#include "SpuContactResult.h"
+
+
+//#define DEBUG_SPU_COLLISION_DETECTION 1
+
+
+#include "SpuContactResult.h"
+
+
+SpuContactResult::SpuContactResult()
+{
+	m_manifoldAddress = 0;
+	m_spuManifold = NULL;
+	m_RequiresWriteBack = false;
+}
+
+ SpuContactResult::~SpuContactResult()
+{
+	g_manifoldDmaExport.swapBuffers();
+}
+
+ void	SpuContactResult::setContactInfo(btPersistentManifold* spuManifold, uint64_t	manifoldAddress,const btTransform& worldTrans0,const btTransform& worldTrans1)
+ {
+//	spu_printf("SpuContactResult::setContactInfo\n");
+	m_rootWorldTransform0 = worldTrans0;
+	m_rootWorldTransform1 = worldTrans1;
+	m_manifoldAddress = manifoldAddress;
+    m_spuManifold = spuManifold;
+ }
+
+ void SpuContactResult::setShapeIdentifiers(int partId0,int index0,	int partId1,int index1)
+ {
+	
+ }
+	
+
+
+ ///return true if it requires a dma transfer back
+bool ManifoldResultAddContactPoint(const btVector3& normalOnBInWorld,
+								   const btVector3& pointInWorld,
+								   float depth,
+								   btPersistentManifold* manifoldPtr,
+								   btTransform& transA,
+								   btTransform& transB
+								   )
+{
+	
+	float contactTreshold = manifoldPtr->getContactBreakingThreshold();
+
+	//spu_printf("SPU: add contactpoint, depth:%f, contactTreshold %f, manifoldPtr %llx\n",depth,contactTreshold,manifoldPtr);
+
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+	spu_printf("SPU: contactTreshold %f\n",contactTreshold);
+#endif //DEBUG_SPU_COLLISION_DETECTION
+	if (depth > manifoldPtr->getContactBreakingThreshold())
+		return false;
+
+	//provide inverses or just calculate?
+	btTransform transAInv = transA.inverse();//m_body0->m_cachedInvertedWorldTransform;
+	btTransform transBInv= transB.inverse();//m_body1->m_cachedInvertedWorldTransform;
+
+	btVector3 pointA = pointInWorld + normalOnBInWorld * depth;
+	btVector3 localA = transAInv(pointA );
+	btVector3 localB = transBInv(pointInWorld);
+	btManifoldPoint newPt(localA,localB,normalOnBInWorld,depth);
+
+	int insertIndex = manifoldPtr->getCacheEntry(newPt);
+	if (insertIndex >= 0)
+	{
+//		manifoldPtr->replaceContactPoint(newPt,insertIndex);
+//		return true;
+
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+		spu_printf("SPU: same contact detected, nothing done\n");
+#endif //DEBUG_SPU_COLLISION_DETECTION
+		// This is not needed, just use the old info! saves a DMA transfer as well
+	} else
+	{
+
+		newPt.m_combinedFriction = 0.25f;//calculateCombinedFriction(m_body0,m_body1);
+		newPt.m_combinedRestitution = 0.0f;//calculateCombinedRestitution(m_body0,m_body1);
+
+		/*
+		//potential TODO: SPU callbacks, either immediate (local on the SPU), or deferred
+		//User can override friction and/or restitution
+		if (gContactAddedCallback &&
+			//and if either of the two bodies requires custom material
+			 ((m_body0->m_collisionFlags & btCollisionObject::customMaterialCallback) ||
+			   (m_body1->m_collisionFlags & btCollisionObject::customMaterialCallback)))
+		{
+			//experimental feature info, for per-triangle material etc.
+			(*gContactAddedCallback)(newPt,m_body0,m_partId0,m_index0,m_body1,m_partId1,m_index1);
+		}
+		*/
+		manifoldPtr->AddManifoldPoint(newPt);
+		return true;
+
+	}
+	return false;
+	
+}
+
+
+void SpuContactResult::writeDoubleBufferedManifold(btPersistentManifold* lsManifold, btPersistentManifold* mmManifold)
+{
+    memcpy(g_manifoldDmaExport.getFront(),lsManifold,sizeof(btPersistentManifold));
+
+    g_manifoldDmaExport.swapBuffers();
+    g_manifoldDmaExport.backBufferDmaPut((uint64_t)mmManifold, sizeof(btPersistentManifold), DMA_TAG(9));
+	// Should there be any kind of wait here?  What if somebody tries to use this tag again?  What if we call this function again really soon?
+	//no, the swapBuffers does the wait
+}
+
+void SpuContactResult::addContactPoint(const btVector3& normalOnBInWorld,const btPoint3& pointInWorld,float depth)
+{
+//	spu_printf("*** SpuContactResult::addContactPoint: depth = %f\n",depth);
+
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+ //   int sman = sizeof(rage::phManifold);
+//	spu_printf("sizeof_manifold = %i\n",sman);
+#endif //DEBUG_SPU_COLLISION_DETECTION
+
+	btPersistentManifold* localManifold = m_spuManifold;
+
+	btVector3	normalB(normalOnBInWorld.getX(),normalOnBInWorld.getY(),normalOnBInWorld.getZ());
+	btVector3	pointWrld(pointInWorld.getX(),pointInWorld.getY(),pointInWorld.getZ());
+
+	//process the contact point
+	const bool retVal = ManifoldResultAddContactPoint(normalB,
+		pointWrld,
+		depth,
+		localManifold,
+		m_rootWorldTransform0,
+		m_rootWorldTransform1
+		);
+	m_RequiresWriteBack = m_RequiresWriteBack || retVal;
+}
+
+void SpuContactResult::flush()
+{
+	if (m_RequiresWriteBack)
+	{
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+		spu_printf("SPU: Start rage::phManifold Write (Put) DMA\n");
+#endif //DEBUG_SPU_COLLISION_DETECTION
+	//	spu_printf("writeDoubleBufferedManifold\n");
+		writeDoubleBufferedManifold(m_spuManifold, (btPersistentManifold*)m_manifoldAddress);
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+		spu_printf("SPU: Finished (Put) DMA\n");
+#endif //DEBUG_SPU_COLLISION_DETECTION
+	}
+	m_spuManifold = NULL;
+	m_RequiresWriteBack = false;
+}
+
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuContactResult.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuContactResult.h
new file mode 100644
index 000000000..5746a02e5
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuContactResult.h
@@ -0,0 +1,101 @@
+#ifndef SPU_CONTACT_RESULT2_H
+#define SPU_CONTACT_RESULT2_H
+
+
+#ifndef WIN32
+#include <stdint.h>
+#endif
+
+
+
+#ifdef WIN32
+#include "SpuDoubleBuffer.h"
+#else
+#include "SPU_Common/SpuDefines.h"
+#include "SPU_Common/SpuDoubleBuffer.h"
+#include <spu_printf.h>
+#endif //WIN32
+
+
+#include "LinearMath/btTransform.h"
+#include "LinearMath/btPoint3.h"
+
+
+#include "BulletCollision/NarrowPhaseCollision/btPersistentManifold.h"
+
+
+struct SpuCollisionPairInput
+{
+	uint64_t m_collisionShapes[2];
+	void*	m_spuCollisionShapes[2];
+
+	uint64_t m_persistentManifoldPtr;
+	btVector3	m_primitiveDimensions0;
+	btVector3	m_primitiveDimensions1;
+	int		m_shapeType0;
+	int		m_shapeType1;	
+	float	m_collisionMargin0;
+	float	m_collisionMargin1;
+
+	btTransform	m_worldTransform0;
+	btTransform m_worldTransform1;
+	
+	bool	m_isSwapped;
+
+
+	
+};
+
+
+struct SpuClosestPointInput
+{
+    SpuClosestPointInput()
+        :m_maximumDistanceSquared(float(1e30)),
+        m_stackAlloc(0)
+    {
+    }
+
+    btTransform m_transformA;
+    btTransform m_transformB;
+    float	m_maximumDistanceSquared;
+    class	btStackAlloc* m_stackAlloc;
+	struct SpuConvexPolyhedronVertexData* m_convexVertexData;
+};
+
+///SpuContactResult exports the contact points using double-buffered DMA transfers, only when needed
+///So when an existing contact point is duplicated, no transfer/refresh is performed.
+class SpuContactResult
+{
+    btTransform		m_rootWorldTransform0;
+	btTransform		m_rootWorldTransform1;
+	uint64_t	m_manifoldAddress;
+
+    btPersistentManifold* m_spuManifold;
+	bool m_RequiresWriteBack;
+
+	DoubleBuffer<btPersistentManifold, 1> g_manifoldDmaExport;
+
+	public:
+		SpuContactResult();
+		virtual ~SpuContactResult();
+
+		btPersistentManifold*	GetSpuManifold() const
+		{
+			return m_spuManifold;
+		}
+
+		virtual void setShapeIdentifiers(int partId0,int index0,	int partId1,int index1);
+
+        void	setContactInfo(btPersistentManifold* spuManifold, uint64_t	manifoldAddress,const btTransform& worldTrans0,const btTransform& worldTrans1);
+
+        void writeDoubleBufferedManifold(btPersistentManifold* lsManifold, btPersistentManifold* mmManifold);
+
+        virtual void addContactPoint(const btVector3& normalOnBInWorld,const btPoint3& pointInWorld,float depth);
+
+		void flush();
+};
+
+
+
+#endif //SPU_CONTACT_RESULT2_H
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuConvexPenetrationDepthSolver.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuConvexPenetrationDepthSolver.h
new file mode 100644
index 000000000..9d8c7ff87
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuConvexPenetrationDepthSolver.h
@@ -0,0 +1,56 @@
+/* [SCE CONFIDENTIAL DOCUMENT]
+ * PLAYSTATION(R)3 SPU Optimized Bullet Physics Library (http://bulletphysics.com)
+ *                Copyright (C) 2007 Sony Computer Entertainment Inc.
+ *                                                All Rights Reserved.
+ */
+
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+
+#ifndef CONVEX_PENETRATION_DEPTH_H
+#define CONVEX_PENETRATION_DEPTH_H
+
+
+
+class btStackAlloc;
+class btIDebugDraw;
+class SpuVoronoiSimplexSolver;
+
+#include <LinearMath/btTransform.h>
+#include <LinearMath/btPoint3.h>
+
+
+///ConvexPenetrationDepthSolver provides an interface for penetration depth calculation.
+class SpuConvexPenetrationDepthSolver
+{
+public:	
+	
+	virtual ~SpuConvexPenetrationDepthSolver() {};
+	virtual bool calcPenDepth( SpuVoronoiSimplexSolver& simplexSolver,
+	        void* convexA,void* convexB,int shapeTypeA, int shapeTypeB, float marginA, float marginB,
+            btTransform& transA,const btTransform& transB,
+			btVector3& v, btPoint3& pa, btPoint3& pb,
+			class btIDebugDraw* debugDraw,btStackAlloc* stackAlloc,
+			struct SpuConvexPolyhedronVertexData* convexVertexData
+			) const = 0;
+
+
+};
+
+
+
+#endif //CONVEX_PENETRATION_DEPTH_H
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp
new file mode 100644
index 000000000..183f0935e
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.cpp
@@ -0,0 +1,399 @@
+
+#include "SpuGatheringCollisionTask.h"
+
+#include "SpuDoubleBuffer.h"
+
+#include "../SpuCollisionTaskProcess.h"
+#include "../SpuGatheringCollisionDispatcher.h" //for SPU_BATCHSIZE_BROADPHASE_PAIRS
+
+#include "BulletCollision/BroadphaseCollision/btBroadphaseProxy.h"
+#include "SpuContactManifoldCollisionAlgorithm.h"
+#include "BulletCollision/CollisionDispatch/btCollisionObject.h"
+#include "SpuContactResult.h"
+#include "BulletCollision/CollisionShapes/btOptimizedBvh.h"
+#include "BulletCollision/CollisionShapes/btTriangleIndexVertexArray.h"
+#include "BulletCollision/CollisionShapes/btConvexShape.h"
+#include "BulletCollision/CollisionShapes/btBvhTriangleMeshShape.h"
+#include "SpuMinkowskiPenetrationDepthSolver.h"
+#include "SpuGjkPairDetector.h"
+#include "SpuVoronoiSimplexSolver.h"
+
+#include "SpuLocalSupport.h" //definition of SpuConvexPolyhedronVertexData
+
+
+#ifdef WIN32
+#define spu_printf printf
+#include <stdio.h>
+#endif
+
+
+//int gNumConvexPoints0=0;
+
+///Make sure no destructors are called on this memory
+struct	CollisionTask_LocalStoreMemory
+{
+
+	DoubleBuffer<unsigned char, MIDPHASE_WORKUNIT_PAGE_SIZE> g_workUnitTaskBuffers;
+	btBroadphasePair	gBroadphasePairs[SPU_BATCHSIZE_BROADPHASE_PAIRS];
+	//SpuContactManifoldCollisionAlgorithm	gSpuContactManifoldAlgo;
+	ATTRIBUTE_ALIGNED16(char	gSpuContactManifoldAlgo[sizeof(SpuContactManifoldCollisionAlgorithm)+128]);
+	SpuContactManifoldCollisionAlgorithm*	getlocalCollisionAlgorithm()
+	{
+		return (SpuContactManifoldCollisionAlgorithm*)&gSpuContactManifoldAlgo;
+
+	}
+	btPersistentManifold	gPersistentManifold;
+	btBroadphaseProxy	gProxy0;
+	btBroadphaseProxy	gProxy1;
+	btCollisionObject	gColObj0;
+	btCollisionObject	gColObj1;
+
+	static const int maxShapeSize = 256;//todo: make some compile-time assert that this is value is larger then sizeof(btCollisionShape)
+
+	ATTRIBUTE_ALIGNED16(char	gCollisionShape0[maxShapeSize]);
+	ATTRIBUTE_ALIGNED16(char	gCollisionShape1[maxShapeSize]);
+
+	ATTRIBUTE_ALIGNED16(btScalar	spuUnscaledVertex[4]);
+	ATTRIBUTE_ALIGNED16(int	spuIndices[16]);
+
+	ATTRIBUTE_ALIGNED16(btOptimizedBvh	gOptimizedBvh);
+	ATTRIBUTE_ALIGNED16(btTriangleIndexVertexArray	gTriangleMeshInterface);
+	///only a single mesh part for now, we can add support for multiple parts, but quantized trees don't support this at the moment 
+	ATTRIBUTE_ALIGNED16(btIndexedMesh	gIndexMesh);
+
+	#define MAX_SPU_SUBTREE_HEADERS 32
+	//1024
+	ATTRIBUTE_ALIGNED16(btBvhSubtreeInfo	gSubtreeHeaders[MAX_SPU_SUBTREE_HEADERS]);
+	ATTRIBUTE_ALIGNED16(btQuantizedBvhNode	gSubtreeNodes[MAX_SUBTREE_SIZE_IN_BYTES/sizeof(btQuantizedBvhNode)]);
+
+	SpuConvexPolyhedronVertexData* convexVertexData;
+
+
+};
+
+
+
+
+void* createCollisionLocalStoreMemory()
+{
+	return new CollisionTask_LocalStoreMemory;
+};
+
+
+
+
+void	ProcessSpuConvexConvexCollision(SpuCollisionPairInput* wuInput, CollisionTask_LocalStoreMemory* lsMemPtr, SpuContactResult& spuContacts)
+{
+
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+	//spu_printf("SPU: ProcessSpuConvexConvexCollision\n");
+#endif //DEBUG_SPU_COLLISION_DETECTION
+	//CollisionShape* shape0 = (CollisionShape*)wuInput->m_collisionShapes[0];
+	//CollisionShape* shape1 = (CollisionShape*)wuInput->m_collisionShapes[1];
+	btPersistentManifold* manifold = (btPersistentManifold*)wuInput->m_persistentManifoldPtr;
+	
+
+
+	bool genericGjk = true;
+
+
+
+	if (genericGjk)
+	{
+		//try generic GJK
+
+		SpuVoronoiSimplexSolver vsSolver;
+		SpuMinkowskiPenetrationDepthSolver	penetrationSolver;
+
+		void* shape0Ptr = wuInput->m_spuCollisionShapes[0];
+		void* shape1Ptr = wuInput->m_spuCollisionShapes[1];
+		int shapeType0 = wuInput->m_shapeType0;
+		int shapeType1 = wuInput->m_shapeType1;
+		float marginA = wuInput->m_collisionMargin0;
+		float marginB = wuInput->m_collisionMargin1;
+		
+		SpuClosestPointInput	cpInput;
+		cpInput.m_convexVertexData = lsMemPtr->convexVertexData;
+		cpInput.m_transformA = wuInput->m_worldTransform0;
+		cpInput.m_transformB = wuInput->m_worldTransform1;
+		float sumMargin = (marginA+marginB);
+		cpInput.m_maximumDistanceSquared = sumMargin * sumMargin;
+
+		uint64_t manifoldAddress = (uint64_t)manifold;
+		btPersistentManifold* spuManifold=&lsMemPtr->gPersistentManifold;
+		spuContacts.setContactInfo(spuManifold,manifoldAddress,wuInput->m_worldTransform0,wuInput->m_worldTransform1);
+	
+
+		SpuGjkPairDetector gjk(shape0Ptr,shape1Ptr,shapeType0,shapeType1,marginA,marginB,&vsSolver,&penetrationSolver);
+		gjk.getClosestPoints(cpInput,spuContacts);//,debugDraw);
+	}
+	
+
+}
+
+
+
+void	processCollisionTask(void* userPtr, void* lsMemPtr)
+{
+
+	SpuGatherAndProcessPairsTaskDesc* taskDescPtr = (SpuGatherAndProcessPairsTaskDesc*)userPtr;
+	SpuGatherAndProcessPairsTaskDesc& taskDesc = *taskDescPtr;
+	CollisionTask_LocalStoreMemory*	colMemPtr = (CollisionTask_LocalStoreMemory*)lsMemPtr;
+	CollisionTask_LocalStoreMemory& lsMem = *(colMemPtr);
+	
+	SpuContactResult spuContacts;
+
+	uint64_t dmaInPtr = taskDesc.inPtr;
+	unsigned int numPages = taskDesc.numPages;
+	unsigned int numOnLastPage = taskDesc.numOnLastPage;
+	
+	// prefetch first set of inputs and wait
+	unsigned int nextNumOnPage = (numPages > 1)? MIDPHASE_NUM_WORKUNITS_PER_PAGE : numOnLastPage;
+	lsMem.g_workUnitTaskBuffers.backBufferDmaGet(dmaInPtr, nextNumOnPage*sizeof(SpuGatherAndProcessWorkUnitInput), DMA_TAG(3));
+	dmaInPtr += MIDPHASE_WORKUNIT_PAGE_SIZE;
+
+	for (unsigned int i = 0; i < numPages; i++)
+	{
+		// wait for back buffer dma and swap buffers
+		unsigned char *inputPtr = lsMem.g_workUnitTaskBuffers.swapBuffers();
+
+		// number on current page is number prefetched last iteration
+		unsigned int numOnPage = nextNumOnPage;
+
+		unsigned int j;
+
+
+		// prefetch next set of inputs
+		if (i < numPages-1)
+		{
+			nextNumOnPage = (i == numPages-2)? numOnLastPage : MIDPHASE_NUM_WORKUNITS_PER_PAGE;
+			lsMem.g_workUnitTaskBuffers.backBufferDmaGet(dmaInPtr, nextNumOnPage*sizeof(SpuGatherAndProcessWorkUnitInput), DMA_TAG(3));
+			dmaInPtr += MIDPHASE_WORKUNIT_PAGE_SIZE;
+		}
+
+		SpuGatherAndProcessWorkUnitInput* wuInputs = reinterpret_cast<SpuGatherAndProcessWorkUnitInput *>(inputPtr);
+
+		for (j = 0; j < numOnPage; j++)
+		{
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+			printMidphaseInput(&wuInputs[j]);
+#endif //DEBUG_SPU_COLLISION_DETECTION
+
+
+			int numPairs = wuInputs[j].m_endIndex - wuInputs[j].m_startIndex;
+
+//			printf("startIndex=%d, endIndex = %d\n",wuInputs[j].m_startIndex,wuInputs[j].m_endIndex);
+
+
+			if (numPairs)
+			{
+			
+				{
+					int dmaSize = numPairs*sizeof(SpuGatherAndProcessPairsTaskDesc);
+					uint64_t	dmaPpuAddress = wuInputs[j].m_pairArrayPtr+wuInputs[j].m_startIndex * sizeof(btBroadphasePair);
+					cellDmaGet(&lsMem.gBroadphasePairs, dmaPpuAddress  , dmaSize, DMA_TAG(1), 0, 0);
+					cellDmaWaitTagStatusAll(DMA_MASK(1));
+				}
+				
+				for (int p=0;p<numPairs;p++)
+				{
+					//for each broadphase pair, do something
+
+					btBroadphasePair& pair = lsMem.gBroadphasePairs[p];
+					int userInfo = int(pair.m_userInfo);
+
+					if (userInfo == 2 && pair.m_algorithm && pair.m_pProxy0 && pair.m_pProxy1)
+					{
+						
+						{
+							int dmaSize = sizeof(SpuContactManifoldCollisionAlgorithm);
+							uint64_t	dmaPpuAddress2 = (uint64_t)pair.m_algorithm;
+							cellDmaGet(&lsMem.gSpuContactManifoldAlgo, dmaPpuAddress2  , dmaSize, DMA_TAG(1), 0, 0);
+							cellDmaWaitTagStatusAll(DMA_MASK(1));
+						}
+
+
+
+						SpuCollisionPairInput collisionPairInput;
+						collisionPairInput.m_persistentManifoldPtr = (uint64_t) lsMem.getlocalCollisionAlgorithm()->getContactManifoldPtr();
+				
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+						spu_printf("SPU: manifoldPtr: %llx",collisionPairInput->m_persistentManifoldPtr);
+#endif //DEBUG_SPU_COLLISION_DETECTION
+
+						{
+							int dmaSize = sizeof(btBroadphaseProxy);
+							uint64_t	dmaPpuAddress2 = (uint64_t)pair.m_pProxy0;
+							cellDmaGet(&lsMem.gProxy0, dmaPpuAddress2  , dmaSize, DMA_TAG(1), 0, 0);
+							cellDmaWaitTagStatusAll(DMA_MASK(1));
+						}
+						{
+							int dmaSize = sizeof(btBroadphaseProxy);
+							uint64_t	dmaPpuAddress2 = (uint64_t)pair.m_pProxy1;
+							cellDmaGet(&lsMem.gProxy1, dmaPpuAddress2  , dmaSize, DMA_TAG(2), 0, 0);
+							cellDmaWaitTagStatusAll(DMA_MASK(2));
+						}
+						
+						//btCollisionObject* colObj0 = (btCollisionObject*)gProxy0.m_clientObject;
+						//btCollisionObject* colObj1 = (btCollisionObject*)gProxy1.m_clientObject;
+
+						{
+							int dmaSize = sizeof(btCollisionObject);
+							uint64_t	dmaPpuAddress2 = (uint64_t)lsMem.gProxy0.m_clientObject;
+							cellDmaGet(&lsMem.gColObj0, dmaPpuAddress2  , dmaSize, DMA_TAG(1), 0, 0);
+							cellDmaWaitTagStatusAll(DMA_MASK(1));
+						}
+						{
+							int dmaSize = sizeof(btCollisionObject);
+							uint64_t	dmaPpuAddress2 = (uint64_t)lsMem.gProxy1.m_clientObject;
+							cellDmaGet(&lsMem.gColObj1, dmaPpuAddress2  , dmaSize, DMA_TAG(2), 0, 0);
+							cellDmaWaitTagStatusAll(DMA_MASK(2));
+						}
+
+												
+						
+						///can wait on the combined DMA_MASK, or dma on the same tag
+
+						collisionPairInput.m_shapeType0 = lsMem.getlocalCollisionAlgorithm()->getShapeType0();
+						collisionPairInput.m_shapeType1 = lsMem.getlocalCollisionAlgorithm()->getShapeType1();
+						collisionPairInput.m_collisionMargin0 = lsMem.getlocalCollisionAlgorithm()->getCollisionMargin0();
+						collisionPairInput.m_collisionMargin1 = lsMem.getlocalCollisionAlgorithm()->getCollisionMargin1();
+
+#ifdef DEBUG_SPU_COLLISION_DETECTION
+						spu_printf("SPU collisionPairInput->m_shapeType0 = %d\n",collisionPairInput->m_shapeType0);
+						spu_printf("SPU collisionPairInput->m_shapeType1 = %d\n",collisionPairInput->m_shapeType1);
+#endif //DEBUG_SPU_COLLISION_DETECTION
+
+						if (1)
+						{
+
+							collisionPairInput.m_worldTransform0 = lsMem.gColObj0.getWorldTransform();
+							collisionPairInput.m_worldTransform1 = lsMem.gColObj1.getWorldTransform();
+
+						
+						
+	#ifdef DEBUG_SPU_COLLISION_DETECTION
+							spu_printf("SPU worldTrans0.origin = (%f,%f,%f)\n",
+								collisionPairInput->m_worldTransform0.getOrigin().getX(),
+								collisionPairInput->m_worldTransform0.getOrigin().getY(),
+								collisionPairInput->m_worldTransform0.getOrigin().getZ());
+
+							spu_printf("SPU worldTrans1.origin = (%f,%f,%f)\n",
+								collisionPairInput->m_worldTransform1.getOrigin().getX(),
+								collisionPairInput->m_worldTransform1.getOrigin().getY(),
+								collisionPairInput->m_worldTransform1.getOrigin().getZ());
+	#endif //DEBUG_SPU_COLLISION_DETECTION
+							
+
+							{
+								int dmaSize = sizeof(btPersistentManifold);
+								uint64_t	dmaPpuAddress2 = collisionPairInput.m_persistentManifoldPtr;
+								cellDmaGet(&lsMem.gPersistentManifold, dmaPpuAddress2  , dmaSize, DMA_TAG(1), 0, 0);
+								cellDmaWaitTagStatusAll(DMA_MASK(1));
+							}
+
+							if (btBroadphaseProxy::isConvex(collisionPairInput.m_shapeType0) 
+								&& btBroadphaseProxy::isConvex(collisionPairInput.m_shapeType1))
+							{
+
+								{
+									int dmaSize = lsMem.maxShapeSize;
+									uint64_t	dmaPpuAddress2 = (uint64_t)lsMem.gColObj0.getCollisionShape();
+									cellDmaGet(lsMem.gCollisionShape0, dmaPpuAddress2  , dmaSize, DMA_TAG(1), 0, 0);
+									cellDmaWaitTagStatusAll(DMA_MASK(1));
+								}
+								{
+									int dmaSize = lsMem.maxShapeSize;
+									uint64_t	dmaPpuAddress2 = (uint64_t)lsMem.gColObj1.getCollisionShape();
+									cellDmaGet(lsMem.gCollisionShape1, dmaPpuAddress2  , dmaSize, DMA_TAG(2), 0, 0);
+									cellDmaWaitTagStatusAll(DMA_MASK(2));
+								}
+
+								btConvexShape* spuConvexShape0 = (btConvexShape*)lsMem.gCollisionShape0;
+								btConvexShape* spuConvexShape1 = (btConvexShape*)lsMem.gCollisionShape1;
+
+								btVector3 dim0 = spuConvexShape0->getImplicitShapeDimensions();
+								btVector3 dim1 = spuConvexShape1->getImplicitShapeDimensions();
+
+								collisionPairInput.m_primitiveDimensions0 = dim0;
+								collisionPairInput.m_primitiveDimensions1 = dim1;
+								collisionPairInput.m_collisionShapes[0] = (uint64_t)lsMem.gColObj0.getCollisionShape();
+								collisionPairInput.m_collisionShapes[1] = (uint64_t)lsMem.gColObj1.getCollisionShape();
+								collisionPairInput.m_spuCollisionShapes[0] = spuConvexShape0;
+								collisionPairInput.m_spuCollisionShapes[1] = spuConvexShape1;
+								ProcessSpuConvexConvexCollision(&collisionPairInput,&lsMem, spuContacts);
+							} else
+							{
+								//a non-convex shape is involved
+
+								bool isSwapped = false;
+								bool handleConvexConcave = false;
+
+								if (btBroadphaseProxy::isConcave(collisionPairInput.m_shapeType0) &&
+									btBroadphaseProxy::isConvex(collisionPairInput.m_shapeType1))
+								{
+									isSwapped = true;
+									spu_printf("SPU convex/concave swapped, unsupported!\n");
+									handleConvexConcave = true;
+								}
+								if (btBroadphaseProxy::isConvex(collisionPairInput.m_shapeType0)&&
+									btBroadphaseProxy::isConcave(collisionPairInput.m_shapeType1))
+								{
+									handleConvexConcave = true;
+								}
+								if (handleConvexConcave && !isSwapped)
+								{
+//									spu_printf("SPU: non-convex detected\n");
+
+									{
+//										uint64_t	dmaPpuAddress2 = (uint64_t)gProxy1.m_clientObject;
+//										spu_printf("SPU: gColObj1 trimesh = %llx\n",dmaPpuAddress2);
+									}
+
+									///dma and initialize the convex object
+									{
+										int dmaSize = lsMem.maxShapeSize;
+										uint64_t	dmaPpuAddress2 = (uint64_t)lsMem.gColObj0.getCollisionShape();
+										cellDmaGet(lsMem.gCollisionShape0, dmaPpuAddress2  , dmaSize, DMA_TAG(1), 0, 0);
+										cellDmaWaitTagStatusAll(DMA_MASK(1));
+									}
+									///dma and initialize the convex object
+									{
+										int dmaSize = lsMem.maxShapeSize;
+										uint64_t	dmaPpuAddress2 = (uint64_t)lsMem.gColObj1.getCollisionShape();
+			//							spu_printf("SPU: trimesh = %llx\n",dmaPpuAddress2);
+										cellDmaGet(lsMem.gCollisionShape1, dmaPpuAddress2  , dmaSize, DMA_TAG(2), 0, 0);
+										cellDmaWaitTagStatusAll(DMA_MASK(2));
+									}
+									btConvexShape* spuConvexShape0 = (btConvexShape*)lsMem.gCollisionShape0;
+									btBvhTriangleMeshShape* trimeshShape = (btBvhTriangleMeshShape*)lsMem.gCollisionShape1;
+
+									btVector3 dim0 = spuConvexShape0->getImplicitShapeDimensions();
+									collisionPairInput.m_primitiveDimensions0 = dim0;
+									collisionPairInput.m_collisionShapes[0] = (uint64_t)lsMem.gColObj0.getCollisionShape();
+									collisionPairInput.m_collisionShapes[1] = (uint64_t)lsMem.gColObj1.getCollisionShape();
+									collisionPairInput.m_spuCollisionShapes[0] = spuConvexShape0;
+									collisionPairInput.m_spuCollisionShapes[1] = trimeshShape;
+								
+									btAssert(0);
+									//ProcessConvexConcaveSpuCollision(&collisionPairInput,spuContacts);
+								}
+
+							}
+
+							  spuContacts.flush();
+
+						}			
+						
+
+
+					}
+
+
+				}
+			}
+		}
+	}
+
+}
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h
new file mode 100644
index 000000000..f96270097
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h
@@ -0,0 +1,11 @@
+
+#ifndef SPU_GATHERING_COLLISION_TASK_H
+#define SPU_GATHERING_COLLISION_TASK_H
+
+struct SpuGatherAndProcessPairsTaskDesc;
+
+void	processCollisionTask(void* userPtr, void* lsMemory);
+
+void*	createCollisionLocalStoreMemory();
+
+#endif //SPU_GATHERING_COLLISION_TASK_H
\ No newline at end of file
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector.cpp
new file mode 100644
index 000000000..cf0022187
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector.cpp
@@ -0,0 +1,310 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "SpuGjkPairDetector.h"
+#include "SpuConvexPenetrationDepthSolver.h"
+#include "SpuLocalSupport.h"
+
+
+#if defined(DEBUG) || defined (_DEBUG)
+#include <stdio.h> //for debug printf
+#ifdef __SPU__
+#include <spu_printf.h>
+#define printf spu_printf
+#endif //__SPU__
+#endif
+
+//must be above the machine epsilon
+#define REL_ERROR2 btScalar(1.0e-6)
+
+//temp globals, to improve GJK/EPA/penetration calculations
+int gSpuNumDeepPenetrationChecks = 0;
+int gSpuNumGjkChecks = 0;
+
+
+
+SpuGjkPairDetector::SpuGjkPairDetector(void* objectA,void* objectB,int shapeTypeA, int shapeTypeB, float marginA,float marginB,SpuVoronoiSimplexSolver* simplexSolver, const SpuConvexPenetrationDepthSolver*	penetrationDepthSolver)
+:m_cachedSeparatingAxis(float(0.),float(0.),float(1.)),
+m_penetrationDepthSolver(penetrationDepthSolver),
+m_simplexSolver(simplexSolver),
+m_minkowskiA(objectA),
+m_minkowskiB(objectB),
+m_shapeTypeA(shapeTypeA),
+m_shapeTypeB(shapeTypeB),
+m_marginA(marginA),
+m_marginB(marginB),
+m_ignoreMargin(false),
+m_lastUsedMethod(-1),
+m_catchDegeneracies(1)
+{
+}
+
+void SpuGjkPairDetector::getClosestPoints(const SpuClosestPointInput& input,SpuContactResult& output)
+{
+	btScalar distance=btScalar(0.);
+	btVector3	normalInB(btScalar(0.),btScalar(0.),btScalar(0.));
+	btVector3 pointOnA,pointOnB;
+	btTransform	localTransA = input.m_transformA;
+	btTransform localTransB = input.m_transformB;
+	btVector3 positionOffset = (localTransA.getOrigin() + localTransB.getOrigin()) * btScalar(0.5);
+	localTransA.getOrigin() -= positionOffset;
+	localTransB.getOrigin() -= positionOffset;
+
+	btScalar marginA = m_marginA;
+	btScalar marginB = m_marginB;
+
+	gSpuNumGjkChecks++;
+
+	//for CCD we don't use margins
+	if (m_ignoreMargin)
+	{
+		marginA = btScalar(0.);
+		marginB = btScalar(0.);
+	}
+
+	m_curIter = 0;
+	int gGjkMaxIter = 1000;//this is to catch invalid input, perhaps check for #NaN?
+	m_cachedSeparatingAxis.setValue(0,1,0);
+
+	bool isValid = false;
+	bool checkSimplex = false;
+	bool checkPenetration = true;
+	m_degenerateSimplex = 0;
+
+	m_lastUsedMethod = -1;
+
+	{
+		btScalar squaredDistance = SIMD_INFINITY;
+		btScalar delta = btScalar(0.);
+		
+		btScalar margin = marginA + marginB;
+		
+		
+
+		m_simplexSolver->reset();
+		
+		for ( ; ; )
+		//while (true)
+		{
+
+			btVector3 seperatingAxisInA = (-m_cachedSeparatingAxis)* input.m_transformA.getBasis();
+			btVector3 seperatingAxisInB = m_cachedSeparatingAxis* input.m_transformB.getBasis();
+
+//			btVector3 pInA = m_minkowskiA->localGetSupportingVertexWithoutMargin(seperatingAxisInA);
+//			btVector3 qInB = m_minkowskiB->localGetSupportingVertexWithoutMargin(seperatingAxisInB);
+
+			btVector3 pInA  = localGetSupportingVertexWithoutMargin(m_shapeTypeA, m_minkowskiA, seperatingAxisInA,input.m_convexVertexData);//, &featureIndexA);
+			btVector3 qInB  = localGetSupportingVertexWithoutMargin(m_shapeTypeB, m_minkowskiB, seperatingAxisInB,input.m_convexVertexData);//, &featureIndexB);
+
+
+			btPoint3  pWorld = localTransA(pInA);	
+			btPoint3  qWorld = localTransB(qInB);
+			
+			btVector3 w	= pWorld - qWorld;
+			delta = m_cachedSeparatingAxis.dot(w);
+
+			// potential exit, they don't overlap
+			if ((delta > btScalar(0.0)) && (delta * delta > squaredDistance * input.m_maximumDistanceSquared)) 
+			{
+				checkPenetration = false;
+				break;
+			}
+
+			//exit 0: the new point is already in the simplex, or we didn't come any closer
+			if (m_simplexSolver->inSimplex(w))
+			{
+				m_degenerateSimplex = 1;
+				checkSimplex = true;
+				break;
+			}
+			// are we getting any closer ?
+			btScalar f0 = squaredDistance - delta;
+			btScalar f1 = squaredDistance * REL_ERROR2;
+
+			if (f0 <= f1)
+			{
+				if (f0 <= btScalar(0.))
+				{
+					m_degenerateSimplex = 2;
+				}
+				checkSimplex = true;
+				break;
+			}
+			//add current vertex to simplex
+			m_simplexSolver->addVertex(w, pWorld, qWorld);
+
+			//calculate the closest point to the origin (update vector v)
+			if (!m_simplexSolver->closest(m_cachedSeparatingAxis))
+			{
+				m_degenerateSimplex = 3;
+				checkSimplex = true;
+				break;
+			}
+
+			btScalar previousSquaredDistance = squaredDistance;
+			squaredDistance = m_cachedSeparatingAxis.length2();
+			
+			//redundant m_simplexSolver->compute_points(pointOnA, pointOnB);
+
+			//are we getting any closer ?
+			if (previousSquaredDistance - squaredDistance <= SIMD_EPSILON * previousSquaredDistance) 
+			{ 
+				m_simplexSolver->backup_closest(m_cachedSeparatingAxis);
+				checkSimplex = true;
+				break;
+			}
+
+			  //degeneracy, this is typically due to invalid/uninitialized worldtransforms for a btCollisionObject   
+              if (m_curIter++ > gGjkMaxIter)   
+              {   
+                      #if defined(DEBUG) || defined (_DEBUG)   
+
+                              printf("SpuGjkPairDetector maxIter exceeded:%i\n",m_curIter);   
+                              printf("sepAxis=(%f,%f,%f), squaredDistance = %f, shapeTypeA=%i,shapeTypeB=%i\n",   
+                              m_cachedSeparatingAxis.getX(),   
+                              m_cachedSeparatingAxis.getY(),   
+                              m_cachedSeparatingAxis.getZ(),   
+                              squaredDistance,   
+							  m_shapeTypeA,   
+                              m_shapeTypeB);
+
+                      #endif   
+                      break;   
+
+              } 
+
+
+			bool check = (!m_simplexSolver->fullSimplex());
+			//bool check = (!m_simplexSolver->fullSimplex() && squaredDistance > SIMD_EPSILON * m_simplexSolver->maxVertex());
+
+			if (!check)
+			{
+				//do we need this backup_closest here ?
+				m_simplexSolver->backup_closest(m_cachedSeparatingAxis);
+				break;
+			}
+		}
+
+		if (checkSimplex)
+		{
+			m_simplexSolver->compute_points(pointOnA, pointOnB);
+			normalInB = pointOnA-pointOnB;
+			btScalar lenSqr = m_cachedSeparatingAxis.length2();
+			//valid normal
+			if (lenSqr < 0.0001)
+			{
+				m_degenerateSimplex = 5;
+			} 
+			if (lenSqr > SIMD_EPSILON*SIMD_EPSILON)
+			{
+				btScalar rlen = btScalar(1.) / btSqrt(lenSqr );
+				normalInB *= rlen; //normalize
+				btScalar s = btSqrt(squaredDistance);
+			
+				btAssert(s > btScalar(0.0));
+				pointOnA -= m_cachedSeparatingAxis * (marginA / s);
+				pointOnB += m_cachedSeparatingAxis * (marginB / s);
+				distance = ((btScalar(1.)/rlen) - margin);
+				isValid = true;
+				
+				m_lastUsedMethod = 1;
+			} else
+			{
+				m_lastUsedMethod = 2;
+			}
+		}
+
+		bool catchDegeneratePenetrationCase = 
+			(m_catchDegeneracies && m_penetrationDepthSolver && m_degenerateSimplex && ((distance+margin) < 0.01));
+
+		//if (checkPenetration && !isValid)
+		if (checkPenetration && (!isValid || catchDegeneratePenetrationCase ))
+		{
+			//penetration case
+		
+			//if there is no way to handle penetrations, bail out
+			if (m_penetrationDepthSolver)
+			{
+				// Penetration depth case.
+				btVector3 tmpPointOnA,tmpPointOnB;
+				
+				gSpuNumDeepPenetrationChecks++;
+
+				bool isValid2 = m_penetrationDepthSolver->calcPenDepth( 
+					*m_simplexSolver, 
+					m_minkowskiA,m_minkowskiB,
+                    m_shapeTypeA, m_shapeTypeB,
+                    marginA, marginB,
+					localTransA,localTransB,
+					m_cachedSeparatingAxis, tmpPointOnA, tmpPointOnB,
+					0,input.m_stackAlloc,input.m_convexVertexData
+					);
+
+				if (isValid2)
+				{
+					btVector3 tmpNormalInB = tmpPointOnB-tmpPointOnA;
+					btScalar lenSqr = tmpNormalInB.length2();
+					if (lenSqr > (SIMD_EPSILON*SIMD_EPSILON))
+					{
+						tmpNormalInB /= btSqrt(lenSqr);
+						btScalar distance2 = -(tmpPointOnA-tmpPointOnB).length();
+						//only replace valid penetrations when the result is deeper (check)
+						if (!isValid || (distance2 < distance))
+						{
+							distance = distance2;
+							pointOnA = tmpPointOnA;
+							pointOnB = tmpPointOnB;
+							normalInB = tmpNormalInB;
+							isValid = true;
+							m_lastUsedMethod = 3;
+						} else
+						{
+							
+						}
+					} else
+					{
+						//isValid = false;
+						m_lastUsedMethod = 4;
+					}
+				} else
+				{
+					m_lastUsedMethod = 5;
+				}
+				
+			}
+		}
+	}
+
+	if (isValid)
+	{
+#ifdef __SPU__
+		//spu_printf("distance\n");
+#endif //__CELLOS_LV2__
+
+
+		output.addContactPoint(
+			normalInB,
+			pointOnB+positionOffset,
+			distance);
+		//printf("gjk add:%f",distance);
+	}
+
+
+}
+
+
+
+
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector.h
new file mode 100644
index 000000000..359de295e
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector.h
@@ -0,0 +1,98 @@
+/* [SCE CONFIDENTIAL DOCUMENT]
+ * PLAYSTATION(R)3 SPU Optimized Bullet Physics Library (http://bulletphysics.com)
+ *                Copyright (C) 2007 Sony Computer Entertainment Inc.
+ *                                                All Rights Reserved.
+ */
+
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+
+
+
+#ifndef SPU_GJK_PAIR_DETECTOR_H
+#define SPU_GJK_PAIR_DETECTOR_H
+
+
+
+#include "SpuContactResult.h"
+
+
+#include "SpuVoronoiSimplexSolver.h"
+class SpuConvexPenetrationDepthSolver;
+
+/// btGjkPairDetector uses GJK to implement the btDiscreteCollisionDetectorInterface
+class SpuGjkPairDetector 
+{
+	
+
+	btVector3	m_cachedSeparatingAxis;
+	const SpuConvexPenetrationDepthSolver*	m_penetrationDepthSolver;
+	SpuVoronoiSimplexSolver* m_simplexSolver;
+	void* m_minkowskiA;
+	void* m_minkowskiB;
+    int m_shapeTypeA;
+    int m_shapeTypeB;
+    float m_marginA;
+    float m_marginB;
+	bool		m_ignoreMargin;
+	
+
+public:
+
+	//some debugging to fix degeneracy problems
+	int			m_lastUsedMethod;
+	int			m_curIter;
+	int			m_degenerateSimplex;
+	int			m_catchDegeneracies;
+
+
+	SpuGjkPairDetector(void* objectA,void* objectB,int m_shapeTypeA, int m_shapeTypeB, float marginA, float marginB, SpuVoronoiSimplexSolver* simplexSolver, const SpuConvexPenetrationDepthSolver*	penetrationDepthSolver);
+	virtual ~SpuGjkPairDetector() {};
+
+	virtual void	getClosestPoints(const SpuClosestPointInput& input,SpuContactResult& output);
+
+	void setMinkowskiA(void* minkA)
+	{
+		m_minkowskiA = minkA;
+	}
+
+	void setMinkowskiB(void* minkB)
+	{
+		m_minkowskiB = minkB;
+	}
+
+	void setCachedSeperatingAxis(const btVector3& seperatingAxis)
+	{
+		m_cachedSeparatingAxis = seperatingAxis;
+	}
+
+	void	setPenetrationDepthSolver(SpuConvexPenetrationDepthSolver*	penetrationDepthSolver)
+	{
+		m_penetrationDepthSolver = penetrationDepthSolver;
+	}
+
+	///don't use setIgnoreMargin, it's for Bullet's internal use
+	void	setIgnoreMargin(bool ignoreMargin)
+	{
+		m_ignoreMargin = ignoreMargin;
+	}
+
+
+};
+
+
+
+#endif //SPU_GJK_PAIR_DETECTOR_H
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector_prev.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector_prev.cpp
new file mode 100644
index 000000000..4f66c3a07
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuGjkPairDetector_prev.cpp
@@ -0,0 +1,332 @@
+/*
+
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+
+#include "SpuGjkPairDetector.h"
+#include "SpuConvexPenetrationDepthSolver.h"
+#include "SpuLocalSupport.h"
+
+//#include "BulletCollision/CollisionShapes/btConvexShape.h"
+
+#if defined(DEBUG) || defined (_DEBUG)
+#include <stdio.h> //for debug printf
+#ifdef __SPU__
+#include <spu_printf.h>
+#define printf spu_printf
+#endif //__SPU__
+#endif
+
+//must be above the machine epsilon
+#define REL_ERROR2 float(1.0e-6)
+
+//temp globals, to improve GJK/EPA/penetration calculations
+int gSpuNumDeepPenetrationChecks = 0;
+int gSpuNumGjkChecks = 0;
+
+
+
+SpuGjkPairDetector::SpuGjkPairDetector(void* objectA,void* objectB,int shapeTypeA, int shapeTypeB, float marginA,float marginB,SpuVoronoiSimplexSolver* simplexSolver, const SpuConvexPenetrationDepthSolver*	penetrationDepthSolver)
+:m_cachedSeparatingAxis(float(0.),float(0.),float(1.)),
+m_penetrationDepthSolver(penetrationDepthSolver),
+m_simplexSolver(simplexSolver),
+m_minkowskiA(objectA),
+m_minkowskiB(objectB),
+m_shapeTypeA(shapeTypeA),
+m_shapeTypeB(shapeTypeB),
+m_marginA(marginA),
+m_marginB(marginB),
+m_ignoreMargin(false),
+m_lastUsedMethod(-1),
+m_catchDegeneracies(1)
+{
+}
+
+void SpuGjkPairDetector::getClosestPoints(const SpuClosestPointInput& input,SpuContactResult& output)
+{
+	float distance=float(0.);
+	Vectormath::Aos::Vector3	normalInB(float(0.),float(0.),float(0.));
+	Vectormath::Aos::Point3 pointOnA,pointOnB;
+	Vectormath::Aos::Transform3 localTransA = input.m_transformA;
+	Vectormath::Aos::Transform3 localTransB = input.m_transformB;
+
+	// World space coordinate 
+    Vectormath::Aos::Vector3 localOriginA = localTransA.getTranslation();
+    Vectormath::Aos::Vector3 localOriginB = localTransB.getTranslation();
+
+	// Average instance position.
+	Vectormath::Aos::Vector3 positionOffset = (localOriginA + localOriginB) * float(0.5);
+
+	// Adjust the instance positions so that they're equidistant from the origin.
+	localTransA.setTranslation(localOriginA - positionOffset);
+	localTransB.setTranslation(localOriginB - positionOffset);
+
+	float marginA = m_marginA;
+	float marginB = m_marginB;
+
+	gSpuNumGjkChecks++;
+
+	//for CCD we don't use margins
+	if (m_ignoreMargin)
+	{
+		marginA = float(0.);
+		marginB = float(0.);
+	}
+
+	m_curIter = 0;
+	int gGjkMaxIter = 1000;//this is to catch invalid input, perhaps check for #NaN?
+    m_cachedSeparatingAxis = Vectormath::Aos::Vector3(0.f,1.f,0.f);
+
+	bool isValid = false;
+	bool checkSimplex = false;
+	bool checkPenetration = true;
+	m_degenerateSimplex = 0;
+
+	m_lastUsedMethod = -1;
+
+	{
+		float squaredDistance = 1e30f;
+		// There's no reason to have this delta declared out here, it's set in the loop before it's used and it's not used outside of the loop.
+		float delta = float(0.);
+		
+		float margin = marginA + marginB;
+
+		m_simplexSolver->reset();
+
+		while (true)
+		{
+			// Get the separating axes into each bound's local space.
+			Vectormath::Aos::Vector3 seperatingAxisInA = orthoInverse(input.m_transformA) * (-m_cachedSeparatingAxis);
+			Vectormath::Aos::Vector3 seperatingAxisInB = orthoInverse(input.m_transformB) * m_cachedSeparatingAxis;
+
+            int shapeTypeA = m_shapeTypeA;
+            int shapeTypeB = m_shapeTypeB;
+//			int featureIndexA = 0, featureIndexB = 0;			// Feature index basically means vertex index.
+			Vectormath::Aos::Point3 pInA = localGetSupportingVertexWithoutMargin(shapeTypeA, m_minkowskiA, seperatingAxisInA);//, &featureIndexA);
+			Vectormath::Aos::Point3 qInB = localGetSupportingVertexWithoutMargin(shapeTypeB, m_minkowskiB, seperatingAxisInB);//, &featureIndexB);
+
+			// These are in a 'translated' world space where the origin of that world space corresponds to 'positionOffset' in the 'real' world space.
+            Vectormath::Aos::Point3  pWorld = localTransA * pInA;	
+			Vectormath::Aos::Point3  qWorld = localTransB * qInB;
+
+            //spu_printf("support point A: %f %f %f\n", pWorld.getX(), pWorld.getY(), pWorld.getZ());
+            //spu_printf("support point B: %f %f %f\n", qWorld.getX(), qWorld.getY(), qWorld.getZ());
+
+			// delta is sort of the distance between the current 'closest points' ...
+			// This seems kind of weird to me since m_cachedSeparatingAxis isn't a unit vector, so I'm not really sure what this signifies.
+			Vectormath::Aos::Vector3 w	= pWorld - qWorld;
+			delta = dot(m_cachedSeparatingAxis, w);
+
+			// potential exit, they don't overlap
+			if ((delta > float(0.0)) && (delta * delta > squaredDistance * input.m_maximumDistanceSquared)) 
+			{
+				checkPenetration = false;
+				break;
+			}
+
+			//exit 0: the new point is already in the simplex, or we didn't come any closer
+			if (m_simplexSolver->inSimplex(w))
+			{
+				m_degenerateSimplex = 1;
+				checkSimplex = true;
+				break;
+			}
+
+			// are we getting any closer ?
+			float f0 = squaredDistance - delta;
+			float f1 = squaredDistance * REL_ERROR2;
+
+			// Are we close enough
+			if (f0 <= f1)
+			{
+				if (f0 <= float(0.))
+				{
+					m_degenerateSimplex = 2;
+				}
+				checkSimplex = true;
+				break;
+			}
+			//add current vertex to simplex
+			m_simplexSolver->addVertex(w, pWorld, qWorld);//, featureIndexA, featureIndexB);
+
+			//calculate the closest point to the origin (update vector v)
+			if (!m_simplexSolver->closest(m_cachedSeparatingAxis))
+			{
+				m_degenerateSimplex = 3;
+				checkSimplex = true;
+				break;
+			}
+
+			float previousSquaredDistance = squaredDistance;
+			squaredDistance = lengthSqr(m_cachedSeparatingAxis);
+			
+			//redundant m_simplexSolver->compute_points(pointOnA, pointOnB);
+
+			//are we getting any closer ?
+			if (previousSquaredDistance - squaredDistance <= FLT_EPSILON * previousSquaredDistance) 
+			{ 
+				m_simplexSolver->backup_closest(m_cachedSeparatingAxis);
+				checkSimplex = true;
+				break;
+			}
+
+			  //degeneracy, this is typically due to invalid/uninitialized worldtransforms for a btCollisionObject   
+              if (m_curIter++ > gGjkMaxIter)   
+              {   
+                      #if defined(DEBUG) || defined (_DEBUG)   
+
+                              printf("SpuGjkPairDetector maxIter exceeded:%i\n",m_curIter);   
+                              printf("sepAxis=(%f,%f,%f), squaredDistance = %f, shapeTypeA=%i,shapeTypeB=%i\n",   
+                              m_cachedSeparatingAxis.getX(),   
+                              m_cachedSeparatingAxis.getY(),   
+                              m_cachedSeparatingAxis.getZ(),   
+                              squaredDistance,   
+                              shapeTypeA,   
+                              shapeTypeB);   
+
+                      #endif   
+                      break;   
+
+              } 
+
+
+			bool check = (!m_simplexSolver->fullSimplex());
+			//bool check = (!m_simplexSolver->fullSimplex() && squaredDistance > FLT_EPSILON * m_simplexSolver->maxVertex());
+
+			if (!check)
+			{
+				//do we need this backup_closest here ?
+				m_simplexSolver->backup_closest(m_cachedSeparatingAxis);
+				break;
+			}
+		}
+
+		if (checkSimplex)
+		{
+			m_simplexSolver->compute_points(pointOnA, pointOnB);
+			normalInB = pointOnA-pointOnB;
+			float lenSqr = lengthSqr(m_cachedSeparatingAxis);
+			//valid normal
+			if (lenSqr < 0.0001)
+			{
+				m_degenerateSimplex = 5;
+			} 
+			if (lenSqr > FLT_EPSILON*FLT_EPSILON)
+			{
+				float rlen = float(1.) / sqrtf(lenSqr );
+				normalInB *= rlen; //normalize
+				float s = sqrtf(squaredDistance);
+			
+				btAssert(s > float(0.0));
+				pointOnA -= m_cachedSeparatingAxis * (marginA / s);
+				pointOnB += m_cachedSeparatingAxis * (marginB / s);
+				distance = ((float(1.)/rlen) - margin);
+				isValid = true;
+				
+				m_lastUsedMethod = 1;
+			} else
+			{
+				m_lastUsedMethod = 2;
+			}
+		}
+
+		bool catchDegeneratePenetrationCase = 
+			(m_catchDegeneracies && m_penetrationDepthSolver && m_degenerateSimplex && ((distance+margin) < 0.01));
+
+		//if (checkPenetration && !isValid)
+		if (checkPenetration && (!isValid || catchDegeneratePenetrationCase ))
+		{
+			//penetration case
+		
+			//if there is no way to handle penetrations, bail out
+			if (m_penetrationDepthSolver)
+			{
+				// Penetration depth case.
+				Vectormath::Aos::Point3 tmpPointOnA,tmpPointOnB;
+				
+//				spu_printf("SPU: deep penetration check\n");
+				gSpuNumDeepPenetrationChecks++;
+
+				bool isValid2 = m_penetrationDepthSolver->calcPenDepth( 
+					*m_simplexSolver, 
+					m_minkowskiA,m_minkowskiB,
+                    m_shapeTypeA, m_shapeTypeB,
+                    marginA, marginB,
+					localTransA,localTransB,
+					m_cachedSeparatingAxis, tmpPointOnA, tmpPointOnB,
+					0,input.m_stackAlloc
+					);
+					
+
+				if (isValid2)
+				{
+					Vectormath::Aos::Vector3 tmpNormalInB = tmpPointOnB-tmpPointOnA;
+					float lenSqr = lengthSqr(tmpNormalInB);
+					if (lenSqr > (FLT_EPSILON*FLT_EPSILON))
+					{
+						tmpNormalInB /= sqrtf(lenSqr);
+						float distance2 = -dist(tmpPointOnA,tmpPointOnB);
+						//only replace valid penetrations when the result is deeper (check)
+						if (!isValid || (distance2 < distance))
+						{
+							distance = distance2;
+							pointOnA = tmpPointOnA;
+							pointOnB = tmpPointOnB;
+							normalInB = tmpNormalInB;
+							isValid = true;
+							m_lastUsedMethod = 3;
+						} else
+						{
+							
+						}
+					} else
+					{
+						//isValid = false;
+						m_lastUsedMethod = 4;
+					}
+				} else
+				{
+					m_lastUsedMethod = 5;
+				}
+				
+			}
+		}
+	}
+
+	if (isValid)
+	{
+#ifdef __SPU__
+		//spu_printf("distance\n");
+#endif //__CELLOS_LV2__
+
+
+		Vectormath::Aos::Point3	tmpPtOnB=pointOnB+positionOffset;
+		Vectormath::Aos::Point3	vmPtOnB(tmpPtOnB.getX(),tmpPtOnB.getY(),tmpPtOnB.getZ());
+		Vectormath::Aos::Vector3	vmNormalOnB(normalInB.getX(),normalInB.getY(),normalInB.getZ());
+
+		output.addContactPoint(
+			vmNormalOnB,
+			vmPtOnB,
+			distance
+            );
+			
+
+		//printf("gjk add:%f",distance);
+	}
+}
+
+
+
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuLocalSupport.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuLocalSupport.h
new file mode 100644
index 000000000..669ec768f
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuLocalSupport.h
@@ -0,0 +1,236 @@
+/* [SCE CONFIDENTIAL DOCUMENT]
+ * PLAYSTATION(R)3 SPU Optimized Bullet Physics Library (http://bulletphysics.com)
+ *                Copyright (C) 2007 Sony Computer Entertainment Inc.
+ *                                                All Rights Reserved.
+ */
+
+
+
+
+#include "BulletCollision/BroadphaseCollision/btBroadphaseProxy.h"
+#include "BulletCollision/CollisionShapes/btConvexShape.h"
+#include "BulletCollision/CollisionShapes/btCylinderShape.h"
+
+
+struct	SpuConvexPolyhedronVertexData
+{
+	void*	gSpuConvexShapePtr0;
+	void*	gSpuConvexShapePtr1;
+	btPoint3* gConvexPoints0;
+	btPoint3* gConvexPoints1;
+	int gNumConvexPoints0;
+	int gNumConvexPoints1;
+};
+
+
+inline btPoint3 localGetSupportingVertexWithoutMargin(int shapeType, void* shape, btVector3 localDir,struct	SpuConvexPolyhedronVertexData* convexVertexData)//, int *featureIndex)
+{
+    switch (shapeType)
+    {
+    case SPHERE_SHAPE_PROXYTYPE:
+        {
+            return btPoint3(0,0,0);
+        }
+	case BOX_SHAPE_PROXYTYPE:
+		{
+//			spu_printf("SPU: getSupport BOX_SHAPE_PROXYTYPE\n");
+			btConvexShape* convexShape = (btConvexShape*)shape;
+			btVector3 halfExtents = convexShape->getImplicitShapeDimensions();
+			float margin = convexShape->getMarginNV();
+			halfExtents -= btVector3(margin,margin,margin);
+			return btPoint3(
+				localDir.getX() < 0.0f ? -halfExtents.x() : halfExtents.x(),
+							localDir.getY() < 0.0f ? -halfExtents.y() : halfExtents.y(),
+							localDir.getZ() < 0.0f ? -halfExtents.z() : halfExtents.z());
+		}
+
+	case TRIANGLE_SHAPE_PROXYTYPE:
+		{
+
+			btVector3 dir(localDir.getX(),localDir.getY(),localDir.getZ());
+			btVector3* vertices = (btVector3*)shape;
+			btVector3 dots(dir.dot(vertices[0]), dir.dot(vertices[1]), dir.dot(vertices[2]));
+	  		btVector3 sup = vertices[dots.maxAxis()];
+			return btPoint3(sup.getX(),sup.getY(),sup.getZ());
+			break;
+		}
+
+	case CYLINDER_SHAPE_PROXYTYPE:
+		{
+			btCylinderShape* cylShape = (btCylinderShape*)shape;
+
+			//mapping of halfextents/dimension onto radius/height depends on how cylinder local orientation is (upAxis)
+
+			btVector3 halfExtents = cylShape->getImplicitShapeDimensions();
+			btVector3 v(localDir.getX(),localDir.getY(),localDir.getZ());
+			
+			int cylinderUpAxis = cylShape->getUpAxis();
+			int XX(1),YY(0),ZZ(2);
+
+			switch (cylinderUpAxis)
+			{
+			case 0:
+				{
+					XX = 1;
+					YY = 0;
+					ZZ = 2;
+					break;
+				}
+			case 1:
+				{
+					XX = 0;
+					YY = 1;
+					ZZ = 2;
+				break;
+				}
+			case 2:
+				{
+					XX = 0;
+					YY = 2;
+					ZZ = 1;
+					break;
+				}
+			default:
+				btAssert(0);
+				//printf("SPU:localGetSupportingVertexWithoutMargin unknown Cylinder up-axis\n");
+			};
+
+			btScalar radius = halfExtents[XX];
+			btScalar halfHeight = halfExtents[cylinderUpAxis];
+
+			btVector3 tmp;
+			btScalar d ;
+
+			btScalar s = btSqrt(v[XX] * v[XX] + v[ZZ] * v[ZZ]);
+			if (s != btScalar(0.0))
+			{
+				d = radius / s;  
+				tmp[XX] = v[XX] * d;
+				tmp[YY] = v[YY] < 0.0 ? -halfHeight : halfHeight;
+				tmp[ZZ] = v[ZZ] * d;
+				return btPoint3(tmp.getX(),tmp.getY(),tmp.getZ());
+			}
+			else
+			{
+				tmp[XX] = radius;
+				tmp[YY] = v[YY] < 0.0 ? -halfHeight : halfHeight;
+				tmp[ZZ] = btScalar(0.0);
+				return btPoint3(tmp.getX(),tmp.getY(),tmp.getZ());
+			}
+		}
+
+	case CAPSULE_SHAPE_PROXYTYPE:
+	{
+		//spu_printf("SPU: todo: getSupport CAPSULE_SHAPE_PROXYTYPE\n");
+		btVector3 vec0(localDir.getX(),localDir.getY(),localDir.getZ());
+
+		btConvexShape* cnvxShape = (btConvexShape*)shape;
+		btVector3 halfExtents = cnvxShape->getImplicitShapeDimensions();
+		btScalar halfHeight = halfExtents.getY();
+		btScalar radius = halfExtents.getX();
+		btVector3 supVec(0,0,0);
+
+		btScalar maxDot(btScalar(-1e30));
+
+		btVector3 vec = vec0;
+		btScalar lenSqr = vec.length2();
+		if (lenSqr < btScalar(0.0001))
+		{
+			vec.setValue(1,0,0);
+		} else
+		{
+			btScalar rlen = btScalar(1.) / btSqrt(lenSqr );
+			vec *= rlen;
+		}
+		btVector3 vtx;
+		btScalar newDot;
+		{
+			btVector3 pos(0,halfHeight,0);
+			vtx = pos +vec*(radius);
+			newDot = vec.dot(vtx);
+			if (newDot > maxDot)
+			{
+				maxDot = newDot;
+				supVec = vtx;
+			}
+		}
+		{
+			btVector3 pos(0,-halfHeight,0);
+			vtx = pos +vec*(radius);
+			newDot = vec.dot(vtx);
+			if (newDot > maxDot)
+			{
+				maxDot = newDot;
+				supVec = vtx;
+			}
+		}
+		return btPoint3(supVec.getX(),supVec.getY(),supVec.getZ());
+		break;
+	};
+
+	case CONVEX_HULL_SHAPE_PROXYTYPE:
+		{
+			//spu_printf("SPU: todo: getSupport CONVEX_HULL_SHAPE_PROXYTYPE\n");
+
+		
+
+			btPoint3* points = 0;
+			int numPoints = 0;
+			if (shape==convexVertexData->gSpuConvexShapePtr0)
+			{
+				points = convexVertexData->gConvexPoints0;
+				numPoints = convexVertexData->gNumConvexPoints0;
+			}
+			if (shape == convexVertexData->gSpuConvexShapePtr1)
+			{
+				points = convexVertexData->gConvexPoints1;
+				numPoints = convexVertexData->gNumConvexPoints1;
+			}
+
+		//	spu_printf("numPoints = %d\n",numPoints);
+
+			btVector3 supVec(btScalar(0.),btScalar(0.),btScalar(0.));
+			btScalar newDot,maxDot = btScalar(-1e30);
+
+			btVector3 vec0(localDir.getX(),localDir.getY(),localDir.getZ());
+			btVector3 vec = vec0;
+			btScalar lenSqr = vec.length2();
+			if (lenSqr < btScalar(0.0001))
+			{
+				vec.setValue(1,0,0);
+			} else
+			{
+				btScalar rlen = btScalar(1.) / btSqrt(lenSqr );
+				vec *= rlen;
+			}
+
+
+			for (int i=0;i<numPoints;i++)
+			{
+				btPoint3 vtx = points[i];// * m_localScaling;
+
+				newDot = vec.dot(vtx);
+				if (newDot > maxDot)
+				{
+					maxDot = newDot;
+					supVec = vtx;
+				}
+			}
+			return btPoint3(supVec.getX(),supVec.getY(),supVec.getZ());
+
+			break;
+		};
+
+    default:
+
+		//spu_printf("SPU:(type %i) missing support function\n",shapeType);
+
+		
+#if __ASSERT
+        spu_printf("localGetSupportingVertexWithoutMargin() - Unsupported bound type: %d.\n", shapeType);
+#endif // __ASSERT
+        return btPoint3(0.f, 0.f, 0.f);
+    }
+}
+
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuMinkowskiPenetrationDepthSolver.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuMinkowskiPenetrationDepthSolver.cpp
new file mode 100644
index 000000000..9cf37c281
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuMinkowskiPenetrationDepthSolver.cpp
@@ -0,0 +1,339 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "SpuMinkowskiPenetrationDepthSolver.h"
+#include "SpuVoronoiSimplexSolver.h"
+#include "SpuGjkPairDetector.h"
+#include "SpuContactResult.h"
+
+
+#include "SpuLocalSupport.h"
+
+#define NUM_UNITSPHERE_POINTS 42
+static btVector3	sPenetrationDirections[NUM_UNITSPHERE_POINTS+MAX_PREFERRED_PENETRATION_DIRECTIONS*2] = 
+{
+btVector3(btScalar(0.000000) , btScalar(-0.000000),btScalar(-1.000000)),
+btVector3(btScalar(0.723608) , btScalar(-0.525725),btScalar(-0.447219)),
+btVector3(btScalar(-0.276388) , btScalar(-0.850649),btScalar(-0.447219)),
+btVector3(btScalar(-0.894426) , btScalar(-0.000000),btScalar(-0.447216)),
+btVector3(btScalar(-0.276388) , btScalar(0.850649),btScalar(-0.447220)),
+btVector3(btScalar(0.723608) , btScalar(0.525725),btScalar(-0.447219)),
+btVector3(btScalar(0.276388) , btScalar(-0.850649),btScalar(0.447220)),
+btVector3(btScalar(-0.723608) , btScalar(-0.525725),btScalar(0.447219)),
+btVector3(btScalar(-0.723608) , btScalar(0.525725),btScalar(0.447219)),
+btVector3(btScalar(0.276388) , btScalar(0.850649),btScalar(0.447219)),
+btVector3(btScalar(0.894426) , btScalar(0.000000),btScalar(0.447216)),
+btVector3(btScalar(-0.000000) , btScalar(0.000000),btScalar(1.000000)),
+btVector3(btScalar(0.425323) , btScalar(-0.309011),btScalar(-0.850654)),
+btVector3(btScalar(-0.162456) , btScalar(-0.499995),btScalar(-0.850654)),
+btVector3(btScalar(0.262869) , btScalar(-0.809012),btScalar(-0.525738)),
+btVector3(btScalar(0.425323) , btScalar(0.309011),btScalar(-0.850654)),
+btVector3(btScalar(0.850648) , btScalar(-0.000000),btScalar(-0.525736)),
+btVector3(btScalar(-0.525730) , btScalar(-0.000000),btScalar(-0.850652)),
+btVector3(btScalar(-0.688190) , btScalar(-0.499997),btScalar(-0.525736)),
+btVector3(btScalar(-0.162456) , btScalar(0.499995),btScalar(-0.850654)),
+btVector3(btScalar(-0.688190) , btScalar(0.499997),btScalar(-0.525736)),
+btVector3(btScalar(0.262869) , btScalar(0.809012),btScalar(-0.525738)),
+btVector3(btScalar(0.951058) , btScalar(0.309013),btScalar(0.000000)),
+btVector3(btScalar(0.951058) , btScalar(-0.309013),btScalar(0.000000)),
+btVector3(btScalar(0.587786) , btScalar(-0.809017),btScalar(0.000000)),
+btVector3(btScalar(0.000000) , btScalar(-1.000000),btScalar(0.000000)),
+btVector3(btScalar(-0.587786) , btScalar(-0.809017),btScalar(0.000000)),
+btVector3(btScalar(-0.951058) , btScalar(-0.309013),btScalar(-0.000000)),
+btVector3(btScalar(-0.951058) , btScalar(0.309013),btScalar(-0.000000)),
+btVector3(btScalar(-0.587786) , btScalar(0.809017),btScalar(-0.000000)),
+btVector3(btScalar(-0.000000) , btScalar(1.000000),btScalar(-0.000000)),
+btVector3(btScalar(0.587786) , btScalar(0.809017),btScalar(-0.000000)),
+btVector3(btScalar(0.688190) , btScalar(-0.499997),btScalar(0.525736)),
+btVector3(btScalar(-0.262869) , btScalar(-0.809012),btScalar(0.525738)),
+btVector3(btScalar(-0.850648) , btScalar(0.000000),btScalar(0.525736)),
+btVector3(btScalar(-0.262869) , btScalar(0.809012),btScalar(0.525738)),
+btVector3(btScalar(0.688190) , btScalar(0.499997),btScalar(0.525736)),
+btVector3(btScalar(0.525730) , btScalar(0.000000),btScalar(0.850652)),
+btVector3(btScalar(0.162456) , btScalar(-0.499995),btScalar(0.850654)),
+btVector3(btScalar(-0.425323) , btScalar(-0.309011),btScalar(0.850654)),
+btVector3(btScalar(-0.425323) , btScalar(0.309011),btScalar(0.850654)),
+btVector3(btScalar(0.162456) , btScalar(0.499995),btScalar(0.850654))
+};
+
+bool SpuMinkowskiPenetrationDepthSolver::calcPenDepth( SpuVoronoiSimplexSolver& simplexSolver,
+	        void* convexA,void* convexB,int shapeTypeA, int shapeTypeB, float marginA, float marginB,
+            btTransform& transA,const btTransform& transB,
+			btVector3& v, btPoint3& pa, btPoint3& pb,
+			class btIDebugDraw* debugDraw,btStackAlloc* stackAlloc,
+			struct SpuConvexPolyhedronVertexData* convexVertexData
+			) const
+{
+
+	(void)stackAlloc;
+	(void)v;
+	
+
+	struct btIntermediateResult : public SpuContactResult
+	{
+
+		btIntermediateResult():m_hasResult(false)
+		{
+		}
+		
+		btVector3 m_normalOnBInWorld;
+		btVector3 m_pointInWorld;
+		btScalar m_depth;
+		bool	m_hasResult;
+
+		virtual void setShapeIdentifiers(int partId0,int index0,	int partId1,int index1)
+		{
+			(void)partId0;
+			(void)index0;
+			(void)partId1;
+			(void)index1;
+		}
+		void addContactPoint(const btVector3& normalOnBInWorld,const btVector3& pointInWorld,btScalar depth)
+		{
+			m_normalOnBInWorld = normalOnBInWorld;
+			m_pointInWorld = pointInWorld;
+			m_depth = depth;
+			m_hasResult = true;
+		}
+	};
+
+	//just take fixed number of orientation, and sample the penetration depth in that direction
+	btScalar minProj = btScalar(1e30);
+	btVector3 minNorm;
+	btVector3 minVertex;
+	btVector3 minA,minB;
+	btVector3 seperatingAxisInA,seperatingAxisInB;
+	btVector3 pInA,qInB,pWorld,qWorld,w;
+
+//#define USE_BATCHED_SUPPORT 1
+#ifdef USE_BATCHED_SUPPORT
+
+	btVector3	supportVerticesABatch[NUM_UNITSPHERE_POINTS+MAX_PREFERRED_PENETRATION_DIRECTIONS*2];
+	btVector3	supportVerticesBBatch[NUM_UNITSPHERE_POINTS+MAX_PREFERRED_PENETRATION_DIRECTIONS*2];
+	btVector3	seperatingAxisInABatch[NUM_UNITSPHERE_POINTS+MAX_PREFERRED_PENETRATION_DIRECTIONS*2];
+	btVector3	seperatingAxisInBBatch[NUM_UNITSPHERE_POINTS+MAX_PREFERRED_PENETRATION_DIRECTIONS*2];
+	int i;
+
+	int numSampleDirections = NUM_UNITSPHERE_POINTS;
+
+	for (i=0;i<numSampleDirections;i++)
+	{
+		const btVector3& norm = sPenetrationDirections[i];
+		seperatingAxisInABatch[i] =  (-norm) * transA.getBasis() ;
+		seperatingAxisInBBatch[i] =  norm   * transB.getBasis() ;
+	}
+
+	{
+		int numPDA = convexA->getNumPreferredPenetrationDirections();
+		if (numPDA)
+		{
+			for (int i=0;i<numPDA;i++)
+			{
+				btVector3 norm;
+				convexA->getPreferredPenetrationDirection(i,norm);
+				norm  = transA.getBasis() * norm;
+				sPenetrationDirections[numSampleDirections] = norm;
+				seperatingAxisInABatch[numSampleDirections] = (-norm) * transA.getBasis();
+				seperatingAxisInBBatch[numSampleDirections] = norm * transB.getBasis();
+				numSampleDirections++;
+			}
+		}
+	}
+
+	{
+		int numPDB = convexB->getNumPreferredPenetrationDirections();
+		if (numPDB)
+		{
+			for (int i=0;i<numPDB;i++)
+			{
+				btVector3 norm;
+				convexB->getPreferredPenetrationDirection(i,norm);
+				norm  = transB.getBasis() * norm;
+				sPenetrationDirections[numSampleDirections] = norm;
+				seperatingAxisInABatch[numSampleDirections] = (-norm) * transA.getBasis();
+				seperatingAxisInBBatch[numSampleDirections] = norm * transB.getBasis();
+				numSampleDirections++;
+			}
+		}
+	}
+
+
+
+	convexA->batchedUnitVectorGetSupportingVertexWithoutMargin(seperatingAxisInABatch,supportVerticesABatch,numSampleDirections);
+	convexB->batchedUnitVectorGetSupportingVertexWithoutMargin(seperatingAxisInBBatch,supportVerticesBBatch,numSampleDirections);
+
+	for (i=0;i<numSampleDirections;i++)
+	{
+		const btVector3& norm = sPenetrationDirections[i];
+		seperatingAxisInA = seperatingAxisInABatch[i];
+		seperatingAxisInB = seperatingAxisInBBatch[i];
+
+		pInA = supportVerticesABatch[i];
+		qInB = supportVerticesBBatch[i];
+
+		pWorld = transA(pInA);	
+		qWorld = transB(qInB);
+		w	= qWorld - pWorld;
+		btScalar delta = norm.dot(w);
+		//find smallest delta
+		if (delta < minProj)
+		{
+			minProj = delta;
+			minNorm = norm;
+			minA = pWorld;
+			minB = qWorld;
+		}
+	}	
+#else
+
+	int numSampleDirections = NUM_UNITSPHERE_POINTS;
+
+#ifdef DO_PREFERRED_DIRECTIONS
+	{
+		int numPDA = convexA->getNumPreferredPenetrationDirections();
+		if (numPDA)
+		{
+			for (int i=0;i<numPDA;i++)
+			{
+				btVector3 norm;
+				convexA->getPreferredPenetrationDirection(i,norm);
+				norm  = transA.getBasis() * norm;
+				sPenetrationDirections[numSampleDirections] = norm;
+				numSampleDirections++;
+			}
+		}
+	}
+
+	{
+		int numPDB = convexB->getNumPreferredPenetrationDirections();
+		if (numPDB)
+		{
+			for (int i=0;i<numPDB;i++)
+			{
+				btVector3 norm;
+				convexB->getPreferredPenetrationDirection(i,norm);
+				norm  = transB.getBasis() * norm;
+				sPenetrationDirections[numSampleDirections] = norm;
+				numSampleDirections++;
+			}
+		}
+	}
+#endif //DO_PREFERRED_DIRECTIONS
+
+	for (int i=0;i<numSampleDirections;i++)
+	{
+		const btVector3& norm = sPenetrationDirections[i];
+		seperatingAxisInA = (-norm)* transA.getBasis();
+		seperatingAxisInB = norm* transB.getBasis();
+
+		pInA = localGetSupportingVertexWithoutMargin(shapeTypeA, convexA, seperatingAxisInA,convexVertexData);//, NULL);
+		qInB = localGetSupportingVertexWithoutMargin(shapeTypeB, convexB, seperatingAxisInB,convexVertexData);//, NULL);
+
+	//	pInA = convexA->localGetSupportingVertexWithoutMargin(seperatingAxisInA);
+	//	qInB = convexB->localGetSupportingVertexWithoutMargin(seperatingAxisInB);
+
+		pWorld = transA(pInA);	
+		qWorld = transB(qInB);
+		w	= qWorld - pWorld;
+		btScalar delta = norm.dot(w);
+		//find smallest delta
+		if (delta < minProj)
+		{
+			minProj = delta;
+			minNorm = norm;
+			minA = pWorld;
+			minB = qWorld;
+		}
+	}
+#endif //USE_BATCHED_SUPPORT
+
+	//add the margins
+
+	minA += minNorm*marginA;
+	minB -= minNorm*marginB;
+	//no penetration
+	if (minProj < btScalar(0.))
+		return false;
+
+	minProj += (marginA + marginB);
+
+
+
+
+
+//#define DEBUG_DRAW 1
+#ifdef DEBUG_DRAW
+	if (debugDraw)
+	{
+		btVector3 color(0,1,0);
+		debugDraw->drawLine(minA,minB,color);
+		color = btVector3 (1,1,1);
+		btVector3 vec = minB-minA;
+		btScalar prj2 = minNorm.dot(vec);
+		debugDraw->drawLine(minA,minA+(minNorm*minProj),color);
+
+	}
+#endif //DEBUG_DRAW
+
+	
+
+	SpuGjkPairDetector gjkdet(convexA,convexB,shapeTypeA,shapeTypeB,marginA,marginB,&simplexSolver,0);
+
+	btScalar offsetDist = minProj;
+	btVector3 offset = minNorm * offsetDist;
+	
+
+	SpuClosestPointInput input;
+
+	btVector3 newOrg = transA.getOrigin() + offset;
+
+	btTransform displacedTrans = transA;
+	displacedTrans.setOrigin(newOrg);
+
+	input.m_transformA = displacedTrans;
+	input.m_transformB = transB;
+	input.m_maximumDistanceSquared = btScalar(1e30);//minProj;
+	
+	btIntermediateResult res;
+	gjkdet.getClosestPoints(input,res);
+
+	btScalar correctedMinNorm = minProj - res.m_depth;
+
+
+	//the penetration depth is over-estimated, relax it
+	btScalar penetration_relaxation= btScalar(1.);
+	minNorm*=penetration_relaxation;
+
+	if (res.m_hasResult)
+	{
+
+		pa = res.m_pointInWorld - minNorm * correctedMinNorm;
+		pb = res.m_pointInWorld;
+		
+#ifdef DEBUG_DRAW
+		if (debugDraw)
+		{
+			btVector3 color(1,0,0);
+			debugDraw->drawLine(pa,pb,color);
+		}
+#endif//DEBUG_DRAW
+
+
+	}
+	return res.m_hasResult;
+}
+
+
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuMinkowskiPenetrationDepthSolver.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuMinkowskiPenetrationDepthSolver.h
new file mode 100644
index 000000000..ffad53b87
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuMinkowskiPenetrationDepthSolver.h
@@ -0,0 +1,51 @@
+/* [SCE CONFIDENTIAL DOCUMENT]
+ * PLAYSTATION(R)3 SPU Optimized Bullet Physics Library (http://bulletphysics.com)
+ *                Copyright (C) 2007 Sony Computer Entertainment Inc.
+ *                                                All Rights Reserved.
+ */
+
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef MINKOWSKI_PENETRATION_DEPTH_SOLVER_H
+#define MINKOWSKI_PENETRATION_DEPTH_SOLVER_H
+
+
+#include "SpuConvexPenetrationDepthSolver.h"
+
+class btStackAlloc;
+class btIDebugDraw;
+class SpuVoronoiSimplexSolver;
+
+///MinkowskiPenetrationDepthSolver implements bruteforce penetration depth estimation.
+///Implementation is based on sampling the depth using support mapping, and using GJK step to get the witness points.
+class SpuMinkowskiPenetrationDepthSolver : public SpuConvexPenetrationDepthSolver
+{
+public:
+
+	virtual bool calcPenDepth( SpuVoronoiSimplexSolver& simplexSolver,
+	        void* convexA,void* convexB,int shapeTypeA, int shapeTypeB, float marginA, float marginB,
+            btTransform& transA,const btTransform& transB,
+			btVector3& v, btPoint3& pa, btPoint3& pb,
+			class btIDebugDraw* debugDraw,btStackAlloc* stackAlloc,
+			struct SpuConvexPolyhedronVertexData* convexVertexData
+			) const;
+
+
+};
+
+
+#endif //MINKOWSKI_PENETRATION_DEPTH_SOLVER_H
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver.cpp
new file mode 100644
index 000000000..a4557dfc4
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver.cpp
@@ -0,0 +1,606 @@
+
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+	
+	Elsevier CDROM license agreements grants nonexclusive license to use the software
+	for any purpose, commercial or non-commercial as long as the following credit is included
+	identifying the original source of the software:
+
+	Parts of the source are "from the book Real-Time Collision Detection by
+	Christer Ericson, published by Morgan Kaufmann Publishers,
+	(c) 2005 Elsevier Inc."
+		
+*/
+
+
+#include "SpuVoronoiSimplexSolver.h"
+#include <assert.h>
+#include <stdio.h>
+
+#define VERTA  0
+#define VERTB  1
+#define VERTC  2
+#define VERTD  3
+
+#define CATCH_DEGENERATE_TETRAHEDRON 1
+void	SpuVoronoiSimplexSolver::removeVertex(int index)
+{
+	
+	assert(m_numVertices>0);
+	m_numVertices--;
+	m_simplexVectorW[index] = m_simplexVectorW[m_numVertices];
+	m_simplexPointsP[index] = m_simplexPointsP[m_numVertices];
+	m_simplexPointsQ[index] = m_simplexPointsQ[m_numVertices];
+}
+
+void	SpuVoronoiSimplexSolver::reduceVertices (const SpuUsageBitfield& usedVerts)
+{
+	if ((numVertices() >= 4) && (!usedVerts.usedVertexD))
+		removeVertex(3);
+
+	if ((numVertices() >= 3) && (!usedVerts.usedVertexC))
+		removeVertex(2);
+
+	if ((numVertices() >= 2) && (!usedVerts.usedVertexB))
+		removeVertex(1);
+	
+	if ((numVertices() >= 1) && (!usedVerts.usedVertexA))
+		removeVertex(0);
+
+}
+
+
+
+
+
+//clear the simplex, remove all the vertices
+void SpuVoronoiSimplexSolver::reset()
+{
+	m_cachedValidClosest = false;
+	m_numVertices = 0;
+	m_needsUpdate = true;
+	m_lastW = btVector3(btScalar(1e30),btScalar(1e30),btScalar(1e30));
+	m_cachedBC.reset();
+}
+
+
+
+	//add a vertex
+void SpuVoronoiSimplexSolver::addVertex(const btVector3& w, const btPoint3& p, const btPoint3& q)
+{
+	m_lastW = w;
+	m_needsUpdate = true;
+
+	m_simplexVectorW[m_numVertices] = w;
+	m_simplexPointsP[m_numVertices] = p;
+	m_simplexPointsQ[m_numVertices] = q;
+
+	m_numVertices++;
+}
+
+bool	SpuVoronoiSimplexSolver::updateClosestVectorAndPoints()
+{
+	
+	if (m_needsUpdate)
+	{
+		m_cachedBC.reset();
+
+		m_needsUpdate = false;
+
+		switch (numVertices())
+		{
+		case 0:
+				m_cachedValidClosest = false;
+				break;
+		case 1:
+			{
+				m_cachedP1 = m_simplexPointsP[0];
+				m_cachedP2 = m_simplexPointsQ[0];
+				m_cachedV = m_cachedP1-m_cachedP2; //== m_simplexVectorW[0]
+				m_cachedBC.reset();
+				m_cachedBC.setBarycentricCoordinates(btScalar(1.),btScalar(0.),btScalar(0.),btScalar(0.));
+				m_cachedValidClosest = m_cachedBC.isValid();
+				break;
+			};
+		case 2:
+			{
+			//closest point origin from line segment
+					const btVector3& from = m_simplexVectorW[0];
+					const btVector3& to = m_simplexVectorW[1];
+					btVector3 nearest;
+
+					btVector3 p (btScalar(0.),btScalar(0.),btScalar(0.));
+					btVector3 diff = p - from;
+					btVector3 v = to - from;
+					btScalar t = v.dot(diff);
+					
+					if (t > 0) {
+						btScalar dotVV = v.dot(v);
+						if (t < dotVV) {
+							t /= dotVV;
+							diff -= t*v;
+							m_cachedBC.m_usedVertices.usedVertexA = true;
+							m_cachedBC.m_usedVertices.usedVertexB = true;
+						} else {
+							t = 1;
+							diff -= v;
+							//reduce to 1 point
+							m_cachedBC.m_usedVertices.usedVertexB = true;
+						}
+					} else
+					{
+						t = 0;
+						//reduce to 1 point
+						m_cachedBC.m_usedVertices.usedVertexA = true;
+					}
+					m_cachedBC.setBarycentricCoordinates(1-t,t);
+					nearest = from + t*v;
+
+					m_cachedP1 = m_simplexPointsP[0] + t * (m_simplexPointsP[1] - m_simplexPointsP[0]);
+					m_cachedP2 = m_simplexPointsQ[0] + t * (m_simplexPointsQ[1] - m_simplexPointsQ[0]);
+					m_cachedV = m_cachedP1 - m_cachedP2;
+					
+					reduceVertices(m_cachedBC.m_usedVertices);
+
+					m_cachedValidClosest = m_cachedBC.isValid();
+					break;
+			}
+		case 3:
+			{
+				//closest point origin from triangle
+				btVector3 p (btScalar(0.),btScalar(0.),btScalar(0.));
+				
+				const btVector3& a = m_simplexVectorW[0];
+				const btVector3& b = m_simplexVectorW[1];
+				const btVector3& c = m_simplexVectorW[2];
+
+				closestPtPointTriangle(p,a,b,c,m_cachedBC);
+				m_cachedP1 = m_simplexPointsP[0] * m_cachedBC.m_barycentricCoords[0] +
+								m_simplexPointsP[1] * m_cachedBC.m_barycentricCoords[1] +
+								m_simplexPointsP[2] * m_cachedBC.m_barycentricCoords[2] +
+								m_simplexPointsP[3] * m_cachedBC.m_barycentricCoords[3];
+
+				m_cachedP2 = m_simplexPointsQ[0] * m_cachedBC.m_barycentricCoords[0] +
+					m_simplexPointsQ[1] * m_cachedBC.m_barycentricCoords[1] +
+					m_simplexPointsQ[2] * m_cachedBC.m_barycentricCoords[2] +
+					m_simplexPointsQ[3] * m_cachedBC.m_barycentricCoords[3];
+
+				m_cachedV = m_cachedP1-m_cachedP2;
+
+				reduceVertices (m_cachedBC.m_usedVertices);
+				m_cachedValidClosest =  m_cachedBC.isValid();
+
+				break;
+			}
+		case 4:
+			{
+
+				
+				btVector3 p (btScalar(0.),btScalar(0.),btScalar(0.));
+				
+				const btVector3& a = m_simplexVectorW[0];
+				const btVector3& b = m_simplexVectorW[1];
+				const btVector3& c = m_simplexVectorW[2];
+				const btVector3& d = m_simplexVectorW[3];
+
+				bool hasSeperation = closestPtPointTetrahedron(p,a,b,c,d,m_cachedBC);
+
+				if (hasSeperation)
+				{
+
+					m_cachedP1 = m_simplexPointsP[0] * m_cachedBC.m_barycentricCoords[0] +
+						m_simplexPointsP[1] * m_cachedBC.m_barycentricCoords[1] +
+						m_simplexPointsP[2] * m_cachedBC.m_barycentricCoords[2] +
+						m_simplexPointsP[3] * m_cachedBC.m_barycentricCoords[3];
+
+					m_cachedP2 = m_simplexPointsQ[0] * m_cachedBC.m_barycentricCoords[0] +
+						m_simplexPointsQ[1] * m_cachedBC.m_barycentricCoords[1] +
+						m_simplexPointsQ[2] * m_cachedBC.m_barycentricCoords[2] +
+						m_simplexPointsQ[3] * m_cachedBC.m_barycentricCoords[3];
+
+					m_cachedV = m_cachedP1-m_cachedP2;
+					reduceVertices (m_cachedBC.m_usedVertices);
+				} else
+				{
+//					printf("sub distance got penetration\n");
+
+					if (m_cachedBC.m_degenerate)
+					{
+						m_cachedValidClosest = false;
+					} else
+					{
+						m_cachedValidClosest = true;
+						//degenerate case == false, penetration = true + zero
+						m_cachedV.setValue(btScalar(0.),btScalar(0.),btScalar(0.));
+					}
+					break;
+				}
+
+				m_cachedValidClosest = m_cachedBC.isValid();
+
+				//closest point origin from tetrahedron
+				break;
+			}
+		default:
+			{
+				m_cachedValidClosest = false;
+			}
+		};
+	}
+
+	return m_cachedValidClosest;
+
+}
+
+//return/calculate the closest vertex
+bool SpuVoronoiSimplexSolver::closest(btVector3& v)
+{
+	bool succes = updateClosestVectorAndPoints();
+	v = m_cachedV;
+	return succes;
+}
+
+
+
+btScalar SpuVoronoiSimplexSolver::maxVertex()
+{
+	int i, numverts = numVertices();
+	btScalar maxV = btScalar(0.);
+	for (i=0;i<numverts;i++)
+	{
+		btScalar curLen2 = m_simplexVectorW[i].length2();
+		if (maxV < curLen2)
+			maxV = curLen2;
+	}
+	return maxV;
+}
+
+
+
+	//return the current simplex
+int SpuVoronoiSimplexSolver::getSimplex(btPoint3 *pBuf, btPoint3 *qBuf, btVector3 *yBuf) const
+{
+	int i;
+	for (i=0;i<numVertices();i++)
+	{
+		yBuf[i] = m_simplexVectorW[i];
+		pBuf[i] = m_simplexPointsP[i];
+		qBuf[i] = m_simplexPointsQ[i];
+	}
+	return numVertices();
+}
+
+
+
+
+bool SpuVoronoiSimplexSolver::inSimplex(const btVector3& w)
+{
+	bool found = false;
+	int i, numverts = numVertices();
+	//btScalar maxV = btScalar(0.);
+	
+	//w is in the current (reduced) simplex
+	for (i=0;i<numverts;i++)
+	{
+		if (m_simplexVectorW[i] == w)
+			found = true;
+	}
+
+	//check in case lastW is already removed
+	if (w == m_lastW)
+		return true;
+    	
+	return found;
+}
+
+void SpuVoronoiSimplexSolver::backup_closest(btVector3& v) 
+{
+	v = m_cachedV;
+}
+
+
+bool SpuVoronoiSimplexSolver::emptySimplex() const 
+{
+	return (numVertices() == 0);
+
+}
+
+void SpuVoronoiSimplexSolver::compute_points(btPoint3& p1, btPoint3& p2) 
+{
+	updateClosestVectorAndPoints();
+	p1 = m_cachedP1;
+	p2 = m_cachedP2;
+
+}
+
+
+
+
+bool	SpuVoronoiSimplexSolver::closestPtPointTriangle(const btPoint3& p, const btPoint3& a, const btPoint3& b, const btPoint3& c,SpuSubSimplexClosestResult& result)
+{
+	result.m_usedVertices.reset();
+
+    // Check if P in vertex region outside A
+    btVector3 ab = b - a;
+    btVector3 ac = c - a;
+    btVector3 ap = p - a;
+    btScalar d1 = ab.dot(ap);
+    btScalar d2 = ac.dot(ap);
+    if (d1 <= btScalar(0.0) && d2 <= btScalar(0.0)) 
+	{
+		result.m_closestPointOnSimplex = a;
+		result.m_usedVertices.usedVertexA = true;
+		result.setBarycentricCoordinates(1,0,0);
+		return true;// a; // barycentric coordinates (1,0,0)
+	}
+
+    // Check if P in vertex region outside B
+    btVector3 bp = p - b;
+    btScalar d3 = ab.dot(bp);
+    btScalar d4 = ac.dot(bp);
+    if (d3 >= btScalar(0.0) && d4 <= d3) 
+	{
+		result.m_closestPointOnSimplex = b;
+		result.m_usedVertices.usedVertexB = true;
+		result.setBarycentricCoordinates(0,1,0);
+
+		return true; // b; // barycentric coordinates (0,1,0)
+	}
+    // Check if P in edge region of AB, if so return projection of P onto AB
+    btScalar vc = d1*d4 - d3*d2;
+    if (vc <= btScalar(0.0) && d1 >= btScalar(0.0) && d3 <= btScalar(0.0)) {
+        btScalar v = d1 / (d1 - d3);
+		result.m_closestPointOnSimplex = a + v * ab;
+		result.m_usedVertices.usedVertexA = true;
+		result.m_usedVertices.usedVertexB = true;
+		result.setBarycentricCoordinates(1-v,v,0);
+		return true;
+        //return a + v * ab; // barycentric coordinates (1-v,v,0)
+    }
+
+    // Check if P in vertex region outside C
+    btVector3 cp = p - c;
+    btScalar d5 = ab.dot(cp);
+    btScalar d6 = ac.dot(cp);
+    if (d6 >= btScalar(0.0) && d5 <= d6) 
+	{
+		result.m_closestPointOnSimplex = c;
+		result.m_usedVertices.usedVertexC = true;
+		result.setBarycentricCoordinates(0,0,1);
+		return true;//c; // barycentric coordinates (0,0,1)
+	}
+
+    // Check if P in edge region of AC, if so return projection of P onto AC
+    btScalar vb = d5*d2 - d1*d6;
+    if (vb <= btScalar(0.0) && d2 >= btScalar(0.0) && d6 <= btScalar(0.0)) {
+        btScalar w = d2 / (d2 - d6);
+		result.m_closestPointOnSimplex = a + w * ac;
+		result.m_usedVertices.usedVertexA = true;
+		result.m_usedVertices.usedVertexC = true;
+		result.setBarycentricCoordinates(1-w,0,w);
+		return true;
+        //return a + w * ac; // barycentric coordinates (1-w,0,w)
+    }
+
+    // Check if P in edge region of BC, if so return projection of P onto BC
+    btScalar va = d3*d6 - d5*d4;
+    if (va <= btScalar(0.0) && (d4 - d3) >= btScalar(0.0) && (d5 - d6) >= btScalar(0.0)) {
+        btScalar w = (d4 - d3) / ((d4 - d3) + (d5 - d6));
+		
+		result.m_closestPointOnSimplex = b + w * (c - b);
+		result.m_usedVertices.usedVertexB = true;
+		result.m_usedVertices.usedVertexC = true;
+		result.setBarycentricCoordinates(0,1-w,w);
+		return true;		
+       // return b + w * (c - b); // barycentric coordinates (0,1-w,w)
+    }
+
+    // P inside face region. Compute Q through its barycentric coordinates (u,v,w)
+    btScalar denom = btScalar(1.0) / (va + vb + vc);
+    btScalar v = vb * denom;
+    btScalar w = vc * denom;
+    
+	result.m_closestPointOnSimplex = a + ab * v + ac * w;
+	result.m_usedVertices.usedVertexA = true;
+	result.m_usedVertices.usedVertexB = true;
+	result.m_usedVertices.usedVertexC = true;
+	result.setBarycentricCoordinates(1-v-w,v,w);
+	
+	return true;
+//	return a + ab * v + ac * w; // = u*a + v*b + w*c, u = va * denom = btScalar(1.0) - v - w
+
+}
+
+
+
+
+
+/// Test if point p and d lie on opposite sides of plane through abc
+int SpuVoronoiSimplexSolver::pointOutsideOfPlane(const btPoint3& p, const btPoint3& a, const btPoint3& b, const btPoint3& c, const btPoint3& d)
+{
+	btVector3 normal = (b-a).cross(c-a);
+
+    btScalar signp = (p - a).dot(normal); // [AP AB AC]
+    btScalar signd = (d - a).dot( normal); // [AD AB AC]
+
+#ifdef CATCH_DEGENERATE_TETRAHEDRON
+#ifdef BT_USE_DOUBLE_PRECISION
+if (signd * signd < (btScalar(1e-8) * btScalar(1e-8)))
+	{
+		return -1;
+	}
+#else
+	if (signd * signd < (btScalar(1e-4) * btScalar(1e-4)))
+	{
+//		printf("affine dependent/degenerate\n");//
+		return -1;
+	}
+#endif
+
+#endif
+	// Points on opposite sides if expression signs are opposite
+    return signp * signd < btScalar(0.);
+}
+
+
+bool	SpuVoronoiSimplexSolver::closestPtPointTetrahedron(const btPoint3& p, const btPoint3& a, const btPoint3& b, const btPoint3& c, const btPoint3& d, SpuSubSimplexClosestResult& finalResult)
+{
+	SpuSubSimplexClosestResult tempResult;
+
+    // Start out assuming point inside all halfspaces, so closest to itself
+	finalResult.m_closestPointOnSimplex = p;
+	finalResult.m_usedVertices.reset();
+    finalResult.m_usedVertices.usedVertexA = true;
+	finalResult.m_usedVertices.usedVertexB = true;
+	finalResult.m_usedVertices.usedVertexC = true;
+	finalResult.m_usedVertices.usedVertexD = true;
+
+    int pointOutsideABC = pointOutsideOfPlane(p, a, b, c, d);
+	int pointOutsideACD = pointOutsideOfPlane(p, a, c, d, b);
+  	int	pointOutsideADB = pointOutsideOfPlane(p, a, d, b, c);
+	int	pointOutsideBDC = pointOutsideOfPlane(p, b, d, c, a);
+
+   if (pointOutsideABC < 0 || pointOutsideACD < 0 || pointOutsideADB < 0 || pointOutsideBDC < 0)
+   {
+	   finalResult.m_degenerate = true;
+	   return false;
+   }
+
+   if (!pointOutsideABC  && !pointOutsideACD && !pointOutsideADB && !pointOutsideBDC)
+	 {
+		 return false;
+	 }
+
+
+    btScalar bestSqDist = FLT_MAX;
+    // If point outside face abc then compute closest point on abc
+	if (pointOutsideABC) 
+	{
+        closestPtPointTriangle(p, a, b, c,tempResult);
+		btPoint3 q = tempResult.m_closestPointOnSimplex;
+		
+        btScalar sqDist = (q - p).dot( q - p);
+        // Update best closest point if (squared) distance is less than current best
+        if (sqDist < bestSqDist) {
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = q;
+			//convert result bitmask!
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexA = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexB = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexC = tempResult.m_usedVertices.usedVertexC;
+			finalResult.setBarycentricCoordinates(
+					tempResult.m_barycentricCoords[VERTA],
+					tempResult.m_barycentricCoords[VERTB],
+					tempResult.m_barycentricCoords[VERTC],
+					0
+			);
+
+		}
+    }
+  
+
+	// Repeat test for face acd
+	if (pointOutsideACD) 
+	{
+        closestPtPointTriangle(p, a, c, d,tempResult);
+		btPoint3 q = tempResult.m_closestPointOnSimplex;
+		//convert result bitmask!
+
+        btScalar sqDist = (q - p).dot( q - p);
+        if (sqDist < bestSqDist) 
+		{
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = q;
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexA = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexC = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexD = tempResult.m_usedVertices.usedVertexC;
+			finalResult.setBarycentricCoordinates(
+					tempResult.m_barycentricCoords[VERTA],
+					0,
+					tempResult.m_barycentricCoords[VERTB],
+					tempResult.m_barycentricCoords[VERTC]
+			);
+
+		}
+    }
+    // Repeat test for face adb
+
+	
+	if (pointOutsideADB)
+	{
+		closestPtPointTriangle(p, a, d, b,tempResult);
+		btPoint3 q = tempResult.m_closestPointOnSimplex;
+		//convert result bitmask!
+
+        btScalar sqDist = (q - p).dot( q - p);
+        if (sqDist < bestSqDist) 
+		{
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = q;
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexA = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexD = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexB = tempResult.m_usedVertices.usedVertexC;
+			finalResult.setBarycentricCoordinates(
+					tempResult.m_barycentricCoords[VERTA],
+					tempResult.m_barycentricCoords[VERTC],
+					0,
+					tempResult.m_barycentricCoords[VERTB]
+			);
+
+		}
+    }
+    // Repeat test for face bdc
+    
+
+	if (pointOutsideBDC)
+	{
+        closestPtPointTriangle(p, b, d, c,tempResult);
+		btPoint3 q = tempResult.m_closestPointOnSimplex;
+		//convert result bitmask!
+        btScalar sqDist = (q - p).dot( q - p);
+        if (sqDist < bestSqDist) 
+		{
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = q;
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexB = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexD = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexC = tempResult.m_usedVertices.usedVertexC;
+
+			finalResult.setBarycentricCoordinates(
+					0,
+					tempResult.m_barycentricCoords[VERTA],
+					tempResult.m_barycentricCoords[VERTC],
+					tempResult.m_barycentricCoords[VERTB]
+			);
+
+		}
+    }
+
+	//help! we ended up full !
+	
+	if (finalResult.m_usedVertices.usedVertexA &&
+		finalResult.m_usedVertices.usedVertexB &&
+		finalResult.m_usedVertices.usedVertexC &&
+		finalResult.m_usedVertices.usedVertexD) 
+	{
+		return true;
+	}
+
+    return true;
+}
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver.h b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver.h
new file mode 100644
index 000000000..cdd3ae6e5
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver.h
@@ -0,0 +1,161 @@
+/* [SCE CONFIDENTIAL DOCUMENT]
+ * PLAYSTATION(R)3 SPU Optimized Bullet Physics Library (http://bulletphysics.com)
+ *                Copyright (C) 2007 Sony Computer Entertainment Inc.
+ *                                                All Rights Reserved.
+ */
+
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+
+
+#ifndef SPUVoronoiSimplexSolver_H
+#define SPUVoronoiSimplexSolver_H
+
+#include <LinearMath/btTransform.h>
+#include <LinearMath/btPoint3.h>
+
+#define VORONOI_SIMPLEX_MAX_VERTS 5
+
+struct SpuUsageBitfield{
+	SpuUsageBitfield()
+	{
+		reset();
+	}
+
+	void reset()
+	{
+		usedVertexA = false;
+		usedVertexB = false;
+		usedVertexC = false;
+		usedVertexD = false;
+	}
+	unsigned short usedVertexA	: 1;
+	unsigned short usedVertexB	: 1;
+	unsigned short usedVertexC	: 1;
+	unsigned short usedVertexD	: 1;
+	unsigned short unused1		: 1;
+	unsigned short unused2		: 1;
+	unsigned short unused3		: 1;
+	unsigned short unused4		: 1;
+};
+
+
+struct	SpuSubSimplexClosestResult
+{
+	btVector3	m_closestPointOnSimplex;
+	//MASK for m_usedVertices
+	//stores the simplex vertex-usage, using the MASK, 
+	// if m_usedVertices & MASK then the related vertex is used
+	SpuUsageBitfield	m_usedVertices;
+	float	m_barycentricCoords[4];
+	bool m_degenerate;
+
+	void	reset()
+	{
+		m_degenerate = false;
+		setBarycentricCoordinates();
+		m_usedVertices.reset();
+	}
+	bool	isValid()
+	{
+		bool valid = (m_barycentricCoords[0] >= float(0.)) &&
+			(m_barycentricCoords[1] >= float(0.)) &&
+			(m_barycentricCoords[2] >= float(0.)) &&
+			(m_barycentricCoords[3] >= float(0.));
+
+
+		return valid;
+	}
+	void	setBarycentricCoordinates(float a=float(0.),float b=float(0.),float c=float(0.),float d=float(0.))
+	{
+		m_barycentricCoords[0] = a;
+		m_barycentricCoords[1] = b;
+		m_barycentricCoords[2] = c;
+		m_barycentricCoords[3] = d;
+	}
+
+};
+
+/// SpuVoronoiSimplexSolver is an implementation of the closest point distance algorithm from a 1-4 points simplex to the origin.
+/// Can be used with GJK, as an alternative to Johnson distance algorithm.
+class SpuVoronoiSimplexSolver
+{
+public:
+
+	int	m_numVertices;
+
+	btVector3	m_simplexVectorW[VORONOI_SIMPLEX_MAX_VERTS];
+	btVector3	m_simplexPointsP[VORONOI_SIMPLEX_MAX_VERTS];
+	btVector3	m_simplexPointsQ[VORONOI_SIMPLEX_MAX_VERTS];
+
+	int m_VertexIndexA[VORONOI_SIMPLEX_MAX_VERTS];
+	int m_VertexIndexB[VORONOI_SIMPLEX_MAX_VERTS];
+
+	btVector3	m_cachedP1;
+	btVector3	m_cachedP2;
+	btVector3	m_cachedV;
+	btVector3	m_lastW;
+	bool		m_cachedValidClosest;
+
+	SpuSubSimplexClosestResult m_cachedBC;
+
+	bool	m_needsUpdate;
+
+	void	removeVertex(int index);
+	void	reduceVertices (const SpuUsageBitfield& usedVerts);
+	bool	updateClosestVectorAndPoints();
+
+	bool	closestPtPointTetrahedron(const btVector3& p, const btVector3& a, const btVector3& b, const btVector3& c, const btVector3& d, SpuSubSimplexClosestResult& finalResult);
+	int		pointOutsideOfPlane(const btVector3& p, const btVector3& a, const btVector3& b, const btVector3& c, const btVector3& d);
+	bool	closestPtPointTriangle(const btVector3& p, const btVector3& a, const btVector3& b, const btVector3& c,SpuSubSimplexClosestResult& result);
+
+	int RemoveDegenerateIndices (const int *inArray, int numIndices, int *outArray) const;
+
+public:
+
+	void reset();
+
+	void addVertex(const btVector3& w, const btPoint3& p, const btPoint3& q);
+
+
+	bool closest(btVector3& v);
+
+	float maxVertex();
+
+	bool fullSimplex() const
+	{
+		return (m_numVertices == 4);
+	}
+
+	int getSimplex(btVector3 *pBuf, btVector3 *qBuf, btVector3 *yBuf) const;
+
+	bool inSimplex(const btVector3& w);
+
+	void backup_closest(btVector3& v) ;
+
+	bool emptySimplex() const ;
+
+	void compute_points(btVector3& p1, btVector3& p2) ;
+
+	int numVertices() const 
+	{
+		return m_numVertices;
+	}
+};
+
+
+
+#endif //SpuVoronoiSimplexSolver
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver_prev.cpp b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver_prev.cpp
new file mode 100644
index 000000000..93cfd97c6
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/SpuVoronoiSimplexSolver_prev.cpp
@@ -0,0 +1,649 @@
+
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2006 Erwin Coumans  http://continuousphysics.com/Bullet/
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+	
+	Elsevier CDROM license agreements grants nonexclusive license to use the software
+	for any purpose, commercial or non-commercial as long as the following credit is included
+	identifying the original source of the software:
+
+	Parts of the source are "from the book Real-Time Collision Detection by
+	Christer Ericson, published by Morgan Kaufmann Publishers,
+	(c) 2005 Elsevier Inc."
+		
+*/
+
+
+// Needed to be able to DMA.
+#ifdef WIN32
+#include "SpuFakeDma.h"
+#else
+#include "SPU_Common/SpuDefines.h"
+#include <cell/spurs/common.h>
+#include <cell/dma.h>
+#endif //WIN32
+
+
+
+#include "SpuVoronoiSimplexSolver.h"
+#include "LinearMath/btScalar.h"
+
+#include <assert.h>
+#include <stdio.h>
+
+
+#define VERTA  0
+#define VERTB  1
+#define VERTC  2
+#define VERTD  3
+
+#define CATCH_DEGENERATE_TETRAHEDRON 1
+void	SpuVoronoiSimplexSolver::removeVertex(int index)
+{
+	assert(m_numVertices>0);
+	m_numVertices--;
+	m_simplexVectorW[index] = m_simplexVectorW[m_numVertices];
+	m_simplexPointsP[index] = m_simplexPointsP[m_numVertices];
+	m_simplexPointsQ[index] = m_simplexPointsQ[m_numVertices];
+//	m_VertexIndexA[index] = m_VertexIndexA[m_numVertices];
+//	m_VertexIndexB[index] = m_VertexIndexB[m_numVertices];
+}
+
+void	SpuVoronoiSimplexSolver::reduceVertices (const SpuUsageBitfield& usedVerts)
+{
+	if ((numVertices() >= 4) && (!usedVerts.usedVertexD))
+		removeVertex(3);
+
+	if ((numVertices() >= 3) && (!usedVerts.usedVertexC))
+		removeVertex(2);
+
+	if ((numVertices() >= 2) && (!usedVerts.usedVertexB))
+		removeVertex(1);
+	
+	if ((numVertices() >= 1) && (!usedVerts.usedVertexA))
+		removeVertex(0);
+
+}
+
+
+
+
+
+//clear the simplex, remove all the vertices
+void SpuVoronoiSimplexSolver::reset()
+{
+	m_cachedValidClosest = false;
+	m_numVertices = 0;
+	m_needsUpdate = true;
+	m_lastW = Vectormath::Aos::Vector3(float(1e30),float(1e30),float(1e30));
+	m_cachedBC.reset();
+}
+
+
+
+	//add a vertex
+void SpuVoronoiSimplexSolver::addVertex(const Vectormath::Aos::Vector3& w, const Vectormath::Aos::Point3& p, const Vectormath::Aos::Point3& q)//, int vertexIndexA, int vertexIndexB)
+{
+	m_lastW = w;
+	m_needsUpdate = true;
+
+	m_simplexVectorW[m_numVertices] = w;
+	m_simplexPointsP[m_numVertices] = Vectormath::Aos::Vector3(p);
+	m_simplexPointsQ[m_numVertices] = Vectormath::Aos::Vector3(q);
+
+	//m_VertexIndexA[m_numVertices] = vertexIndexA;
+	//m_VertexIndexB[m_numVertices] = vertexIndexB;
+
+	m_numVertices++;
+}
+
+bool	SpuVoronoiSimplexSolver::updateClosestVectorAndPoints()
+{
+	
+	if (m_needsUpdate)
+	{
+		m_cachedBC.reset();
+
+		m_needsUpdate = false;
+
+		switch (numVertices())
+		{
+		case 0:
+				m_cachedValidClosest = false;
+				break;
+		case 1:
+			{
+				m_cachedP1 = m_simplexPointsP[0];
+				m_cachedP2 = m_simplexPointsQ[0];
+				m_cachedV = m_cachedP1-m_cachedP2; //== m_simplexVectorW[0]
+				m_cachedBC.reset();
+				m_cachedBC.setBarycentricCoordinates(float(1.),float(0.),float(0.),float(0.));
+				m_cachedValidClosest = m_cachedBC.isValid();
+				break;
+			};
+		case 2:
+			{
+			//closest point origin from line segment
+					const Vectormath::Aos::Vector3& from = m_simplexVectorW[0];
+					const Vectormath::Aos::Vector3& to = m_simplexVectorW[1];
+					Vectormath::Aos::Vector3 nearest;
+
+					Vectormath::Aos::Vector3 p (float(0.),float(0.),float(0.));
+					Vectormath::Aos::Vector3 diff = p - from;
+					Vectormath::Aos::Vector3 v = to - from;
+					float t = dot(v, diff);
+					
+					if (t > 0) {
+						float dotVV = dot(v, v);
+						if (t < dotVV) {
+							t /= dotVV;
+							diff -= t*v;
+							m_cachedBC.m_usedVertices.usedVertexA = true;
+							m_cachedBC.m_usedVertices.usedVertexB = true;
+						} else {
+							t = 1;
+							diff -= v;
+							//reduce to 1 point
+							m_cachedBC.m_usedVertices.usedVertexB = true;
+						}
+					} else
+					{
+						t = 0;
+						//reduce to 1 point
+						m_cachedBC.m_usedVertices.usedVertexA = true;
+					}
+					m_cachedBC.setBarycentricCoordinates(1-t,t);
+					nearest = from + t*v;
+
+					m_cachedP1 = m_simplexPointsP[0] + t * (m_simplexPointsP[1] - m_simplexPointsP[0]);
+					m_cachedP2 = m_simplexPointsQ[0] + t * (m_simplexPointsQ[1] - m_simplexPointsQ[0]);
+					m_cachedV = m_cachedP1 - m_cachedP2;
+
+					reduceVertices(m_cachedBC.m_usedVertices);
+
+					m_cachedValidClosest = m_cachedBC.isValid();
+					break;
+			}
+		case 3:
+			{
+				//closest point origin from triangle
+				Vectormath::Aos::Vector3 p (float(0.),float(0.),float(0.));
+				
+				const Vectormath::Aos::Vector3& a = m_simplexVectorW[0];
+				const Vectormath::Aos::Vector3& b = m_simplexVectorW[1];
+				const Vectormath::Aos::Vector3& c = m_simplexVectorW[2];
+
+				closestPtPointTriangle(p,a,b,c,m_cachedBC);
+				m_cachedP1 = m_simplexPointsP[0] * m_cachedBC.m_barycentricCoords[0] +
+								m_simplexPointsP[1] * m_cachedBC.m_barycentricCoords[1] +
+								m_simplexPointsP[2] * m_cachedBC.m_barycentricCoords[2];
+
+				m_cachedP2 = m_simplexPointsQ[0] * m_cachedBC.m_barycentricCoords[0] +
+					m_simplexPointsQ[1] * m_cachedBC.m_barycentricCoords[1] +
+					m_simplexPointsQ[2] * m_cachedBC.m_barycentricCoords[2];
+
+				m_cachedV = m_cachedP1-m_cachedP2;
+
+				reduceVertices (m_cachedBC.m_usedVertices);
+				m_cachedValidClosest =  m_cachedBC.isValid();
+
+				break;
+			}
+		case 4:
+			{
+
+				
+				Vectormath::Aos::Vector3 p (float(0.),float(0.),float(0.));
+				
+				const Vectormath::Aos::Vector3& a = m_simplexVectorW[0];
+				const Vectormath::Aos::Vector3& b = m_simplexVectorW[1];
+				const Vectormath::Aos::Vector3& c = m_simplexVectorW[2];
+				const Vectormath::Aos::Vector3& d = m_simplexVectorW[3];
+
+				bool hasSeperation = closestPtPointTetrahedron(p,a,b,c,d,m_cachedBC);
+
+				if (hasSeperation)
+				{
+
+					m_cachedP1 = m_simplexPointsP[0] * m_cachedBC.m_barycentricCoords[0] +
+						m_simplexPointsP[1] * m_cachedBC.m_barycentricCoords[1] +
+						m_simplexPointsP[2] * m_cachedBC.m_barycentricCoords[2] +
+						m_simplexPointsP[3] * m_cachedBC.m_barycentricCoords[3];
+
+					m_cachedP2 = m_simplexPointsQ[0] * m_cachedBC.m_barycentricCoords[0] +
+						m_simplexPointsQ[1] * m_cachedBC.m_barycentricCoords[1] +
+						m_simplexPointsQ[2] * m_cachedBC.m_barycentricCoords[2] +
+						m_simplexPointsQ[3] * m_cachedBC.m_barycentricCoords[3];
+
+					m_cachedV = m_cachedP1-m_cachedP2;
+					reduceVertices (m_cachedBC.m_usedVertices);
+				} else
+				{
+//					printf("sub distance got penetration\n");
+
+					if (m_cachedBC.m_degenerate)
+					{
+						m_cachedValidClosest = false;
+					} else
+					{
+						m_cachedValidClosest = true;
+						//degenerate case == false, penetration = true + zero
+                        m_cachedV = Vectormath::Aos::Vector3(float(0.),float(0.),float(0.));
+					}
+					break;
+				}
+
+				m_cachedValidClosest = m_cachedBC.isValid();
+
+				//closest point origin from tetrahedron
+				break;
+			}
+		default:
+			{
+				m_cachedValidClosest = false;
+			}
+		};
+	}
+
+	return m_cachedValidClosest;
+
+}
+
+//return/calculate the closest vertex
+bool SpuVoronoiSimplexSolver::closest(Vectormath::Aos::Vector3& v)
+{
+	bool succes = updateClosestVectorAndPoints();
+	v = m_cachedV;
+	return succes;
+}
+
+
+
+float SpuVoronoiSimplexSolver::maxVertex()
+{
+	int i, numverts = numVertices();
+	float maxV = float(0.);
+	for (i=0;i<numverts;i++)
+	{
+		float curLen2 = lengthSqr(m_simplexVectorW[i]);
+		if (maxV < curLen2)
+			maxV = curLen2;
+	}
+	return maxV;
+}
+
+
+
+	//return the current simplex
+int SpuVoronoiSimplexSolver::getSimplex(Vectormath::Aos::Vector3 *pBuf, Vectormath::Aos::Vector3 *qBuf, Vectormath::Aos::Vector3 *yBuf) const
+{
+	int i;
+	for (i=0;i<numVertices();i++)
+	{
+		yBuf[i] = m_simplexVectorW[i];
+		pBuf[i] = m_simplexPointsP[i];
+		qBuf[i] = m_simplexPointsQ[i];
+	}
+	return numVertices();
+}
+
+
+
+
+bool SpuVoronoiSimplexSolver::inSimplex(const Vectormath::Aos::Vector3& w)
+{
+	bool found = false;
+	int i, numverts = numVertices();
+	//float maxV = float(0.);
+	
+	//w is in the current (reduced) simplex
+	for (i=0;i<numverts;i++)
+	{
+        // TODO: find a better way to determine equality
+		if (m_simplexVectorW[i].getX() == w.getX() &&
+            m_simplexVectorW[i].getY() == w.getY() &&
+            m_simplexVectorW[i].getZ() == w.getZ())
+			found = true;
+	}
+
+	//check in case lastW is already removed
+        // TODO: find a better way to determine equality
+	if (w.getX() == m_lastW.getX() &&
+        w.getY() == m_lastW.getY() &&
+        w.getZ() == m_lastW.getZ())
+		return true;
+    	
+	return found;
+}
+
+void SpuVoronoiSimplexSolver::backup_closest(Vectormath::Aos::Vector3& v) 
+{
+	v = m_cachedV;
+}
+
+
+bool SpuVoronoiSimplexSolver::emptySimplex() const 
+{
+	return (numVertices() == 0);
+
+}
+
+void SpuVoronoiSimplexSolver::compute_points(Vectormath::Aos::Point3& p1, Vectormath::Aos::Point3& p2) 
+{
+	updateClosestVectorAndPoints();
+	p1 = Vectormath::Aos::Point3(m_cachedP1);
+	p2 = Vectormath::Aos::Point3(m_cachedP2);
+
+}
+
+
+
+
+bool	SpuVoronoiSimplexSolver::closestPtPointTriangle(const Vectormath::Aos::Vector3& p, const Vectormath::Aos::Vector3& a, const Vectormath::Aos::Vector3& b, const Vectormath::Aos::Vector3& c,SpuSubSimplexClosestResult& result)
+{
+	result.m_usedVertices.reset();
+
+    // Check if P in vertex region outside A
+    Vectormath::Aos::Vector3 ab = b - a;
+    Vectormath::Aos::Vector3 ac = c - a;
+    Vectormath::Aos::Vector3 ap = p - a;
+    float d1 = dot(ab,ap);
+    float d2 = dot(ac,ap);
+    if (d1 <= float(0.0) && d2 <= float(0.0)) 
+	{
+		result.m_closestPointOnSimplex = Vectormath::Aos::Point3(a);
+		result.m_usedVertices.usedVertexA = true;
+		result.setBarycentricCoordinates(1,0,0);
+		return true;// a; // barycentric coordinates (1,0,0)
+	}
+
+    // Check if P in vertex region outside B
+    Vectormath::Aos::Vector3 bp = p - b;
+    float d3 = dot(ab,bp);
+    float d4 = dot(ac,bp);
+    if (d3 >= float(0.0) && d4 <= d3) 
+	{
+		result.m_closestPointOnSimplex = Vectormath::Aos::Point3(b);
+		result.m_usedVertices.usedVertexB = true;
+		result.setBarycentricCoordinates(0,1,0);
+
+		return true; // b; // barycentric coordinates (0,1,0)
+	}
+    // Check if P in edge region of AB, if so return projection of P onto AB
+    float vc = d1*d4 - d3*d2;
+    if (vc <= float(0.0) && d1 >= float(0.0) && d3 <= float(0.0)) {
+        float v = d1 / (d1 - d3);
+		result.m_closestPointOnSimplex = Vectormath::Aos::Point3(a + v * ab);
+		result.m_usedVertices.usedVertexA = true;
+		result.m_usedVertices.usedVertexB = true;
+		result.setBarycentricCoordinates(1-v,v,0);
+		return true;
+        //return a + v * ab; // barycentric coordinates (1-v,v,0)
+    }
+
+    // Check if P in vertex region outside C
+    Vectormath::Aos::Vector3 cp = p - c;
+    float d5 = dot(ab,cp);
+    float d6 = dot(ac,cp);
+    if (d6 >= float(0.0) && d5 <= d6) 
+	{
+		result.m_closestPointOnSimplex = Vectormath::Aos::Point3(c);
+		result.m_usedVertices.usedVertexC = true;
+		result.setBarycentricCoordinates(0,0,1);
+		return true;//c; // barycentric coordinates (0,0,1)
+	}
+
+    // Check if P in edge region of AC, if so return projection of P onto AC
+    float vb = d5*d2 - d1*d6;
+    if (vb <= float(0.0) && d2 >= float(0.0) && d6 <= float(0.0)) {
+        float w = d2 / (d2 - d6);
+		result.m_closestPointOnSimplex = Vectormath::Aos::Point3(a + w * ac);
+		result.m_usedVertices.usedVertexA = true;
+		result.m_usedVertices.usedVertexC = true;
+		result.setBarycentricCoordinates(1-w,0,w);
+		return true;
+        //return a + w * ac; // barycentric coordinates (1-w,0,w)
+    }
+
+    // Check if P in edge region of BC, if so return projection of P onto BC
+    float va = d3*d6 - d5*d4;
+    if (va <= float(0.0) && (d4 - d3) >= float(0.0) && (d5 - d6) >= float(0.0)) {
+        float w = (d4 - d3) / ((d4 - d3) + (d5 - d6));
+		
+		result.m_closestPointOnSimplex = Vectormath::Aos::Point3(b + w * (c - b));
+		result.m_usedVertices.usedVertexB = true;
+		result.m_usedVertices.usedVertexC = true;
+		result.setBarycentricCoordinates(0,1-w,w);
+		return true;		
+       // return b + w * (c - b); // barycentric coordinates (0,1-w,w)
+    }
+
+    // P inside face region. Compute Q through its barycentric coordinates (u,v,w)
+    float denom = float(1.0) / (va + vb + vc);
+    float v = vb * denom;
+    float w = vc * denom;
+    
+	result.m_closestPointOnSimplex = Vectormath::Aos::Point3(a + ab * v + ac * w);
+	result.m_usedVertices.usedVertexA = true;
+	result.m_usedVertices.usedVertexB = true;
+	result.m_usedVertices.usedVertexC = true;
+	result.setBarycentricCoordinates(1-v-w,v,w);
+	
+	return true;
+//	return a + ab * v + ac * w; // = u*a + v*b + w*c, u = va * denom = float(1.0) - v - w
+}
+
+
+// This is specifically just removing duplicate indices.
+int SpuVoronoiSimplexSolver::RemoveDegenerateIndices (const int* inArray, int numIndices, int* outArray) const
+{
+	int outIndex = 0;
+	for (int firstIndex=0; firstIndex<numIndices; firstIndex++)
+	{
+		bool duplicate = false;
+		for (int secondIndex=0; secondIndex<firstIndex; secondIndex++)
+		{
+			if (inArray[secondIndex]==inArray[firstIndex])
+			{
+				duplicate = true;
+				break;
+			}
+		}
+
+		if (!duplicate)
+		{
+			outArray[outIndex++] = inArray[firstIndex];
+		}
+	}
+
+	return outIndex;
+}
+
+
+
+
+/// Test if point p and d lie on opposite sides of plane through abc
+int SpuVoronoiSimplexSolver::pointOutsideOfPlane(const Vectormath::Aos::Vector3& p, const Vectormath::Aos::Vector3& a, const Vectormath::Aos::Vector3& b, const Vectormath::Aos::Vector3& c, const Vectormath::Aos::Vector3& d)
+{
+	Vectormath::Aos::Vector3 normal = cross(b-a,c-a);
+
+    float signp = dot(p - a, normal); // [AP AB AC]
+    float signd = dot(d - a, normal); // [AD AB AC]
+
+#ifdef CATCH_DEGENERATE_TETRAHEDRON
+	if (signd * signd < (float(1e-4) * float(1e-4)))
+	{
+//		printf("affine dependent/degenerate\n");//
+		return -1;
+	}
+#endif
+	// Points on opposite sides if expression signs are opposite
+    return signp * signd < float(0.);
+}
+
+
+bool	SpuVoronoiSimplexSolver::closestPtPointTetrahedron(const Vectormath::Aos::Vector3& p, const Vectormath::Aos::Vector3& a, const Vectormath::Aos::Vector3& b, const Vectormath::Aos::Vector3& c, const Vectormath::Aos::Vector3& d, SpuSubSimplexClosestResult& finalResult)
+{
+	SpuSubSimplexClosestResult tempResult;
+
+    // Start out assuming point inside all halfspaces, so closest to itself
+	finalResult.m_closestPointOnSimplex = Vectormath::Aos::Point3(p);
+	finalResult.m_usedVertices.reset();
+    finalResult.m_usedVertices.usedVertexA = true;
+	finalResult.m_usedVertices.usedVertexB = true;
+	finalResult.m_usedVertices.usedVertexC = true;
+	finalResult.m_usedVertices.usedVertexD = true;
+
+	// Check only the tetrahedron faces that are closest to p.  We do that by checking each face (which itself is a triangle) to see if the excluded
+	//   vertex (the one vertex that isn't part of that face) is on the other side of that face from the point p.
+    int pointOutsideABC = pointOutsideOfPlane(p, a, b, c, d);
+	int pointOutsideACD = pointOutsideOfPlane(p, a, c, d, b);
+  	int	pointOutsideADB = pointOutsideOfPlane(p, a, d, b, c);
+	int	pointOutsideBDC = pointOutsideOfPlane(p, b, d, c, a);
+
+   if (pointOutsideABC < 0 || pointOutsideACD < 0 || pointOutsideADB < 0 || pointOutsideBDC < 0)
+   {
+	   finalResult.m_degenerate = true;
+	   return false;
+   }
+
+   if (!pointOutsideABC  && !pointOutsideACD && !pointOutsideADB && !pointOutsideBDC)
+	 {
+		 return false;
+	 }
+
+
+    float bestSqDist = 1e30f;//FLT_MAX;
+    // If point outside face abc then compute closest point on abc
+	if (pointOutsideABC) 
+	{
+        closestPtPointTriangle(p, a, b, c,tempResult);
+        Vectormath::Aos::Vector3 q = Vectormath::Aos::Vector3(tempResult.m_closestPointOnSimplex);
+
+        float sqDist = dot(q - p, q - p);
+        // Update best closest point if (squared) distance is less than current best
+        //if (sqDist < bestSqDist)
+		btAssert(sqDist < bestSqDist);		// This has to be true; we haven't actually tested any other combinations.
+		{
+			bestSqDist = sqDist;
+            finalResult.m_closestPointOnSimplex = Vectormath::Aos::Point3(q);
+			//convert result bitmask!
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexA = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexB = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexC = tempResult.m_usedVertices.usedVertexC;
+			finalResult.setBarycentricCoordinates(
+					tempResult.m_barycentricCoords[VERTA],
+					tempResult.m_barycentricCoords[VERTB],
+					tempResult.m_barycentricCoords[VERTC],
+					0
+			);
+
+		}
+    }
+  
+
+	// Repeat test for face acd
+	if (pointOutsideACD) 
+	{
+        closestPtPointTriangle(p, a, c, d,tempResult);
+		Vectormath::Aos::Vector3 q = Vectormath::Aos::Vector3(tempResult.m_closestPointOnSimplex);
+		//convert result bitmask!
+
+        float sqDist = dot(q - p, q - p);
+        if (sqDist < bestSqDist) 
+		{
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = Vectormath::Aos::Point3(q);
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexA = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexC = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexD = tempResult.m_usedVertices.usedVertexC;
+			finalResult.setBarycentricCoordinates(
+					tempResult.m_barycentricCoords[VERTA],
+					0,
+					tempResult.m_barycentricCoords[VERTB],
+					tempResult.m_barycentricCoords[VERTC]
+			);
+
+		}
+    }
+    // Repeat test for face adb
+
+	
+	if (pointOutsideADB)
+	{
+		closestPtPointTriangle(p, a, d, b,tempResult);
+		Vectormath::Aos::Vector3 q = Vectormath::Aos::Vector3(tempResult.m_closestPointOnSimplex);
+		//convert result bitmask!
+
+        float sqDist = dot(q - p, q - p);
+        if (sqDist < bestSqDist) 
+		{
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = Vectormath::Aos::Point3(q);
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexA = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexD = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexB = tempResult.m_usedVertices.usedVertexC;
+			finalResult.setBarycentricCoordinates(
+					tempResult.m_barycentricCoords[VERTA],
+					tempResult.m_barycentricCoords[VERTC],
+					0,
+					tempResult.m_barycentricCoords[VERTB]
+			);
+
+		}
+    }
+    // Repeat test for face bdc
+    
+
+	if (pointOutsideBDC)
+	{
+        closestPtPointTriangle(p, b, d, c,tempResult);
+		Vectormath::Aos::Vector3 q = Vectormath::Aos::Vector3(tempResult.m_closestPointOnSimplex);
+		//convert result bitmask!
+        float sqDist = dot(q - p, q - p);
+        if (sqDist < bestSqDist) 
+		{
+			bestSqDist = sqDist;
+			finalResult.m_closestPointOnSimplex = Vectormath::Aos::Point3(q);
+			finalResult.m_usedVertices.reset();
+			finalResult.m_usedVertices.usedVertexB = tempResult.m_usedVertices.usedVertexA;
+			finalResult.m_usedVertices.usedVertexD = tempResult.m_usedVertices.usedVertexB;
+			finalResult.m_usedVertices.usedVertexC = tempResult.m_usedVertices.usedVertexC;
+
+			finalResult.setBarycentricCoordinates(
+					0,
+					tempResult.m_barycentricCoords[VERTA],
+					tempResult.m_barycentricCoords[VERTC],
+					tempResult.m_barycentricCoords[VERTB]
+			);
+
+		}
+    }
+
+	//help! we ended up full !
+	
+	if (finalResult.m_usedVertices.usedVertexA &&
+		finalResult.m_usedVertices.usedVertexB &&
+		finalResult.m_usedVertices.usedVertexC &&
+		finalResult.m_usedVertices.usedVertexD) 
+	{
+		return true;
+	}
+
+    return true;
+}
+
+
diff --git a/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/readme.txt b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/readme.txt
new file mode 100644
index 000000000..f3e907347
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuNarrowPhaseCollisionTask/readme.txt
@@ -0,0 +1 @@
+Empty placeholder for future Libspe2 SPU task
diff --git a/Extras/BulletMultiThreaded/SpuSampleTask/readme.txt b/Extras/BulletMultiThreaded/SpuSampleTask/readme.txt
new file mode 100644
index 000000000..f3e907347
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuSampleTask/readme.txt
@@ -0,0 +1 @@
+Empty placeholder for future Libspe2 SPU task
diff --git a/Extras/BulletMultiThreaded/SpuSampleTaskProcess.cpp b/Extras/BulletMultiThreaded/SpuSampleTaskProcess.cpp
new file mode 100644
index 000000000..ca2fff17b
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuSampleTaskProcess.cpp
@@ -0,0 +1,199 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+//#define __CELLOS_LV2__ 1
+
+
+
+#include "SpuLibspe2Support.h"
+#include "Win32ThreadSupport.h"
+
+//#include "SPUAssert.h"
+#include <string.h>
+
+
+#include "SpuSampleTaskProcess.h"
+
+
+#include <stdio.h>
+
+
+void	SampleThreadFunc(void* userPtr,void* lsMemory)
+{
+	//do nothing
+	printf("hello world\n");
+}
+
+void*	SamplelsMemoryFunc()
+{
+	//don't create local store memory, just return 0
+	return 0;
+}
+
+
+//SpuLibspe2Support gSampleSPU(SPU_ELF_SAMPLE,SAMPLE_NUM_WORKUNIT_TASKS);
+Win32ThreadSupport gSampleSPU(Win32ThreadSupport::Win32ThreadConstructionInfo("sample",
+							  		SampleThreadFunc,
+									SamplelsMemoryFunc,
+									SAMPLE_NUM_WORKUNIT_TASKS));
+
+
+extern "C" {
+	extern char SPU_SAMPLE_ELF_SYMBOL[];
+};
+
+
+
+
+
+SpuSampleTaskDesc g_spuSampleTaskDesc[SAMPLE_NUM_WORKUNIT_TASKS];
+
+
+
+SpuSampleTaskProcess::SpuSampleTaskProcess(Win32ThreadSupport::Win32ThreadConstructionInfo& threadConstructionInfo)
+{
+
+	for (int i = 0; i < threadConstructionInfo.m_numThreads; i++)
+	{
+		m_taskBusy[i] = false;
+	}
+	m_numBusyTasks = 0;
+	m_currentTask = 0;
+
+	m_initialized = false;
+
+	gSampleSPU.startSPUs(threadConstructionInfo);
+
+}
+
+SpuSampleTaskProcess::~SpuSampleTaskProcess()
+{
+	
+	gSampleSPU.stopSPUs();
+	
+}
+
+
+
+void	SpuSampleTaskProcess::initialize()
+{
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("SpuSampleTaskProcess::initialize()\n");
+#endif //DEBUG_SPU_TASK_SCHEDULING
+	
+	for (int i = 0; i < SAMPLE_NUM_WORKUNIT_TASKS; i++)
+	{
+		m_taskBusy[i] = false;
+	}
+	m_numBusyTasks = 0;
+	m_currentTask = 0;
+	m_initialized = true;
+
+}
+
+
+void SpuSampleTaskProcess::issueTask(void* sampleMainMemPtr,int sampleValue)
+{
+
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("SpuSampleTaskProcess::issueTask (m_currentTask= %d\)n", m_currentTask);
+#endif //DEBUG_SPU_TASK_SCHEDULING
+
+	m_taskBusy[m_currentTask] = true;
+	m_numBusyTasks++;
+
+	SpuSampleTaskDesc& taskDesc = g_spuSampleTaskDesc[m_currentTask];
+	{
+		// send task description in event message
+		// no error checking here...
+		// but, currently, event queue can be no larger than NUM_WORKUNIT_TASKS.
+	
+		taskDesc.m_mainMemoryPtr = reinterpret_cast<uint64_t>(sampleMainMemPtr);
+		taskDesc.m_sampleValue = sampleValue;
+
+		//some bookkeeping to recognize finished tasks
+		taskDesc.m_taskId = m_currentTask;
+	}
+
+
+	gSampleSPU.sendRequest(CMD_SAMPLE_TASK_COMMAND, (uint32_t) &taskDesc, m_currentTask);
+
+	// if all tasks busy, wait for spu event to clear the task.
+	
+	if (m_numBusyTasks >= SAMPLE_NUM_WORKUNIT_TASKS)
+	{
+		unsigned int taskId;
+		unsigned int outputSize;
+
+		gSampleSPU.waitForResponse(&taskId, &outputSize);
+
+		//printf("PPU: after issue, received event: %u %d\n", taskId, outputSize);
+
+		postProcess(taskId, outputSize);
+
+		m_taskBusy[taskId] = false;
+
+		m_numBusyTasks--;
+	}
+
+	// find new task buffer
+	for (unsigned int i = 0; i < SAMPLE_NUM_WORKUNIT_TASKS; i++)
+	{
+		if (!m_taskBusy[i])
+		{
+			m_currentTask = i;
+			break;
+		}
+	}
+}
+
+
+///Optional PPU-size post processing for each task
+void SpuSampleTaskProcess::postProcess(int taskId, int outputSize)
+{
+
+}
+
+
+void SpuSampleTaskProcess::flush()
+{
+#ifdef DEBUG_SPU_TASK_SCHEDULING
+	printf("\nSpuCollisionTaskProcess::flush()\n");
+#endif //DEBUG_SPU_TASK_SCHEDULING
+	
+
+	// all tasks are issued, wait for all tasks to be complete
+	while(m_numBusyTasks > 0)
+	{
+// Consolidating SPU code
+	  unsigned int taskId;
+	  unsigned int outputSize;
+	  
+	  {
+			
+		  gSampleSPU.waitForResponse(&taskId, &outputSize);
+	  }
+
+		//printf("PPU: flushing, received event: %u %d\n", taskId, outputSize);
+
+		postProcess(taskId, outputSize);
+
+		m_taskBusy[taskId] = false;
+
+		m_numBusyTasks--;
+	}
+
+
+}
diff --git a/Extras/BulletMultiThreaded/SpuSampleTaskProcess.h b/Extras/BulletMultiThreaded/SpuSampleTaskProcess.h
new file mode 100644
index 000000000..243eddc32
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuSampleTaskProcess.h
@@ -0,0 +1,83 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#ifndef SPU_SAMPLE_TASK_PROCESS_H
+#define SPU_SAMPLE_TASK_PROCESS_H
+
+#include <assert.h>
+
+#include <LinearMath/btScalar.h>
+
+#include "PlatformDefinitions.h"
+
+#include <stdlib.h>
+
+///maximum outstanding tasks
+#define SAMPLE_NUM_WORKUNIT_TASKS 4
+
+
+
+///SpuSampleTaskDesc
+struct SpuSampleTaskDesc
+{
+	uint64_t 	m_mainMemoryPtr;
+	int			m_sampleValue;
+	
+	uint16_t 	m_taskId;
+}
+#ifdef __CELLOS_LV2__
+__attribute__ ((aligned (16)))
+#endif
+;
+
+//just add your commands here, try to keep them globally unique for debugging purposes
+#define CMD_SAMPLE_TASK_COMMAND 10
+
+
+
+/// SpuSampleTaskProcess handles SPU processing of collision pairs.
+/// When PPU issues a task, it will look for completed task buffers
+/// PPU will do postprocessing, dependent on workunit output (not likely)
+class SpuSampleTaskProcess
+{
+	// track task buffers that are being used, and total busy tasks
+	bool           m_taskBusy[SAMPLE_NUM_WORKUNIT_TASKS];
+	unsigned int   m_numBusyTasks;
+
+	// the current task and the current entry to insert a new work unit
+	unsigned int   m_currentTask;
+
+	bool m_initialized;
+
+	void postProcess(int taskId, int outputSize);
+	
+
+public:
+	SpuSampleTaskProcess(Win32ThreadSupport::Win32ThreadConstructionInfo& ci);
+	
+	~SpuSampleTaskProcess();
+	
+	///call initialize in the beginning of the frame, before addCollisionPairToTask
+	void initialize();
+
+	void issueTask(void* sampleMainMemPtr,int sampleValue);
+
+	///call flush to submit potential outstanding work to SPUs and wait for all involved SPUs to be finished
+	void flush();
+};
+
+
+#endif // SPU_SAMPLE_TASK_PROCESS_H
+
diff --git a/Extras/BulletMultiThreaded/SpuSolverTask/readme.txt b/Extras/BulletMultiThreaded/SpuSolverTask/readme.txt
new file mode 100644
index 000000000..f3e907347
--- /dev/null
+++ b/Extras/BulletMultiThreaded/SpuSolverTask/readme.txt
@@ -0,0 +1 @@
+Empty placeholder for future Libspe2 SPU task
diff --git a/Extras/BulletMultiThreaded/Win32ThreadSupport.cpp b/Extras/BulletMultiThreaded/Win32ThreadSupport.cpp
new file mode 100644
index 000000000..b0e26aa6a
--- /dev/null
+++ b/Extras/BulletMultiThreaded/Win32ThreadSupport.cpp
@@ -0,0 +1,226 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "Win32ThreadSupport.h"
+
+#ifdef USE_WIN32_THREADING
+
+#include "SpuCollisionTaskProcess.h"
+#include "SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h"
+
+#include <Windows.h>
+
+///Win32ThreadSupport helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
+///Setup and initialize SPU/CELL/Libspe2
+Win32ThreadSupport::Win32ThreadSupport(Win32ThreadConstructionInfo& threadConstructionInfo)
+{
+	startSPUs(threadConstructionInfo);
+}
+
+///cleanup/shutdown Libspe2
+Win32ThreadSupport::~Win32ThreadSupport()
+{
+	stopSPUs();
+}
+
+
+
+
+#include <stdio.h>
+
+DWORD WINAPI Thread_no_1( LPVOID lpParam ) 
+{
+
+	btSpuStatus* status = (btSpuStatus*)lpParam;
+
+	
+	while (1)
+	{
+		WaitForSingleObject(status->m_eventStartHandle,INFINITE);
+		btAssert(status->m_status);
+
+		void* userPtr = status->m_userPtr;
+
+		if (userPtr)
+		{
+			status->m_userThreadFunc(userPtr,status->m_lsMemory);
+			SetEvent(status->m_eventCompletetHandle);
+		} else
+		{
+			//exit Thread
+			break;
+		}
+		
+	}
+
+	return 0;
+
+}
+
+///send messages to SPUs
+void Win32ThreadSupport::sendRequest(uint32_t uiCommand, uint32_t uiArgument0, uint32_t taskId)
+{
+	///	gMidphaseSPU.sendRequest(CMD_GATHER_AND_PROCESS_PAIRLIST, (uint32_t) &taskDesc);
+	
+	///we should spawn an SPU task here, and in 'waitForResponse' it should wait for response of the (one of) the first tasks that finished
+	
+
+
+	switch (uiCommand)
+	{
+	case 	CMD_GATHER_AND_PROCESS_PAIRLIST:
+		{
+
+
+//#define SINGLE_THREADED 1
+#ifdef SINGLE_THREADED
+
+			btSpuStatus&	spuStatus = m_activeSpuStatus[0];
+			spuStatus.m_userPtr=(void*)uiArgument0;
+			spuStatus.m_userThreadFunc(spuStatus.m_userPtr,spuStatus.m_lsMemory);
+			HANDLE handle =0;
+#else
+
+
+			btSpuStatus&	spuStatus = m_activeSpuStatus[taskId];
+			btAssert(taskId>=0);
+			btAssert(taskId<m_activeSpuStatus.size());
+
+			spuStatus.m_commandId = uiCommand;
+			spuStatus.m_status = 1;
+			spuStatus.m_userPtr = (void*)uiArgument0;
+
+			///fire event to start new task
+			SetEvent(spuStatus.m_eventStartHandle);
+
+#endif //CollisionTask_LocalStoreMemory
+
+			
+
+			break;
+		}
+	default:
+		{
+			///not implemented
+			btAssert(0);
+		}
+
+	};
+
+
+}
+
+///check for messages from SPUs
+void Win32ThreadSupport::waitForResponse(unsigned int *puiArgument0, unsigned int *puiArgument1)
+{
+	///We should wait for (one of) the first tasks to finish (or other SPU messages), and report its response
+	
+	///A possible response can be 'yes, SPU handled it', or 'no, please do a PPU fallback'
+
+
+	btAssert(m_activeSpuStatus.size());
+
+	int last = -1;
+	
+	//find an active spu/thread
+	for (int i=0;i<m_activeSpuStatus.size();i++)
+	{
+		if (m_activeSpuStatus[i].m_status)
+		{
+			last = i;
+			break;
+		}
+	}
+
+
+#ifndef SINGLE_THREADED
+	btSpuStatus& spuStatus = m_activeSpuStatus[last];
+	btAssert(spuStatus.m_threadHandle);
+	btAssert(spuStatus.m_eventCompletetHandle);
+
+	WaitForSingleObject(spuStatus.m_eventCompletetHandle, INFINITE);
+	spuStatus.m_status = 0;
+
+	///need to find an active spu
+	btAssert(last>=0);
+
+#else
+	last=0;
+	btSpuStatus& spuStatus = m_activeSpuStatus[last];
+#endif //SINGLE_THREADED
+
+	
+
+	*puiArgument0 = spuStatus.m_taskId;
+	*puiArgument1 = spuStatus.m_status;
+
+
+}
+
+
+///start the spus group (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+void Win32ThreadSupport::startSPUs(Win32ThreadConstructionInfo& threadConstructionInfo)
+{
+
+	m_activeSpuStatus.resize(threadConstructionInfo.m_numThreads);
+
+	for (int i=0;i<threadConstructionInfo.m_numThreads;i++)
+	{
+		printf("starting thread %d\n",i);
+
+		btSpuStatus&	spuStatus = m_activeSpuStatus[i];
+
+		LPSECURITY_ATTRIBUTES lpThreadAttributes=NULL;
+		SIZE_T dwStackSize=threadConstructionInfo.m_threadStackSize;
+		LPTHREAD_START_ROUTINE lpStartAddress=&Thread_no_1;
+		LPVOID lpParameter=&spuStatus;
+		DWORD dwCreationFlags=0;
+		LPDWORD lpThreadId=0;
+
+		spuStatus.m_userPtr=0;
+
+		sprintf(spuStatus.m_eventStartHandleName,"eventStart%s%d",threadConstructionInfo.m_uniqueName,i);
+		spuStatus.m_eventStartHandle = CreateEvent(0,false,false,spuStatus.m_eventStartHandleName);
+
+		sprintf(spuStatus.m_eventCompletetHandleName,"eventComplete%s%d",threadConstructionInfo.m_uniqueName,i);
+		spuStatus.m_eventCompletetHandle = CreateEvent(0,false,false,spuStatus.m_eventCompletetHandleName);
+
+
+		HANDLE handle = CreateThread(lpThreadAttributes,dwStackSize,lpStartAddress,lpParameter,	dwCreationFlags,lpThreadId);
+		SetThreadPriority(handle,THREAD_PRIORITY_TIME_CRITICAL);
+
+		spuStatus.m_taskId = i;
+		spuStatus.m_commandId = 0;
+		spuStatus.m_status = 0;
+		spuStatus.m_threadHandle = handle;
+		spuStatus.m_lsMemory = threadConstructionInfo.m_lsMemoryFunc();
+		spuStatus.m_userThreadFunc = threadConstructionInfo.m_userThreadFunc;
+
+		printf("started thread %d with threadHandle %d\n",i,handle);
+		
+	}
+
+}
+
+///tell the task scheduler we are done with the SPU tasks
+void Win32ThreadSupport::stopSPUs()
+{
+//	m_activeSpuStatus.pop_back();
+//	WaitForSingleObject(spuStatus.bla, INFINITE);
+//	CloseHandle(spuStatus.m_threadHandle);
+	
+}
+
+#endif //USE_WIN32_THREADING
diff --git a/Extras/BulletMultiThreaded/Win32ThreadSupport.h b/Extras/BulletMultiThreaded/Win32ThreadSupport.h
new file mode 100644
index 000000000..70d11d962
--- /dev/null
+++ b/Extras/BulletMultiThreaded/Win32ThreadSupport.h
@@ -0,0 +1,116 @@
+/*
+Bullet Continuous Collision Detection and Physics Library
+Copyright (c) 2003-2007 Erwin Coumans  http://bulletphysics.com
+
+This software is provided 'as-is', without any express or implied warranty.
+In no event will the authors be held liable for any damages arising from the use of this software.
+Permission is granted to anyone to use this software for any purpose, 
+including commercial applications, and to alter it and redistribute it freely, 
+subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+*/
+
+#include "LinearMath/btScalar.h"
+#include "PlatformDefinitions.h"
+
+#ifdef USE_WIN32_THREADING  //platform specific defines are defined in PlatformDefinitions.h
+
+#ifndef WIN32_THREAD_SUPPORT_H
+#define WIN32_THREAD_SUPPORT_H
+
+#include "LinearMath/btAlignedObjectArray.h"
+
+
+
+
+#include <LinearMath/btScalar.h> //for uint32_t etc.
+
+
+typedef void (*Win32ThreadFunc)(void* userPtr,void* lsMemory);
+typedef void* (*Win32lsMemorySetupFunc)();
+
+
+///placeholder, until libspe2 support is there
+struct	btSpuStatus
+{
+	uint32_t	m_taskId;
+	uint32_t	m_commandId;
+	uint32_t	m_status;
+
+	Win32ThreadFunc	m_userThreadFunc;
+	void*	m_userPtr; //for taskDesc etc
+	void*	m_lsMemory; //initialized using Win32LocalStoreMemorySetupFunc
+
+	void*	m_threadHandle; //this one is calling 'Win32ThreadFunc'
+
+	void*	m_eventStartHandle;
+	char	m_eventStartHandleName[32];
+
+	void*	m_eventCompletetHandle;
+	char	m_eventCompletetHandleName[32];
+	
+
+};
+
+
+
+///Win32ThreadSupport helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
+class Win32ThreadSupport {
+
+	btAlignedObjectArray<btSpuStatus>	m_activeSpuStatus;
+
+public:
+	///Setup and initialize SPU/CELL/Libspe2
+
+	
+
+	struct	Win32ThreadConstructionInfo
+	{
+		Win32ThreadConstructionInfo(char* uniqueName,
+									Win32ThreadFunc userThreadFunc,
+									Win32lsMemorySetupFunc	lsMemoryFunc,
+									int numThreads=1,
+									int threadStackSize=65535
+									)
+									:m_uniqueName(uniqueName),
+									m_userThreadFunc(userThreadFunc),
+									m_lsMemoryFunc(lsMemoryFunc),
+									m_numThreads(numThreads),
+									m_threadStackSize(threadStackSize)
+		{
+
+		}
+
+		char*					m_uniqueName;
+		Win32ThreadFunc			m_userThreadFunc;
+		Win32lsMemorySetupFunc	m_lsMemoryFunc;
+		int						m_numThreads;
+		int						m_threadStackSize;
+
+	};
+
+	Win32ThreadSupport(Win32ThreadConstructionInfo& threadConstructionInfo);
+
+///cleanup/shutdown Libspe2
+	~Win32ThreadSupport();
+
+///send messages to SPUs
+	void sendRequest(uint32_t uiCommand, uint32_t uiArgument0, uint32_t uiArgument1);
+
+///check for messages from SPUs
+	void waitForResponse(unsigned int *puiArgument0, unsigned int *puiArgument1);
+
+///start the spus (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+	void startSPUs(Win32ThreadConstructionInfo& threadConstructionInfo);
+
+///tell the task scheduler we are done with the SPU tasks
+	void stopSPUs();
+
+};
+
+#endif //WIN32_THREAD_SUPPORT_H
+
+#endif //USE_WIN32_THREADING